diff --git a/configs/det/dbnet/README.md b/configs/det/dbnet/README.md index 4577ef19b..a6b3d153c 100644 --- a/configs/det/dbnet/README.md +++ b/configs/det/dbnet/README.md @@ -70,14 +70,14 @@ Here we present general purpose models that were trained on wide variety of task The models were trained on 12 public datasets (CTW, LSVT, RCTW-17, TextOCR, etc.) that contain wide range of images. The training set has 153,511 images and the validation set has 9,786 images.
The test set consists of 598 images manually selected from the above-mentioned datasets. -Performance tested on ascend 910 with graph mode +Experiments are tested on ascend 910 with mindspore 2.3.1 graph mode
-| **Model** | **Cards** | **Backbone** | **Languages** | **F-score** | **Batch Size** | **graph compile** | **jit level** | **Step Time** | **Recipe** | **Download** | -|-----------|---------|--------------|-------------------|:---------------------------:|----------------|:-----------------:|-----------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------|----------------|----------------------------------------------------------------------------------------------------------| -| DBNet | 8 | ResNet-50 | Chinese + English | 83.41% | 10 | 107.91 s | O2| 312.48 ms/step | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141-912f0a90.mindir) | -| DBNet++ | 4 | ResNet-50 | Chinese + English | 84.30% | 32 | 182.94 s | O2| 1230.76 ms/step | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9-b3f52398.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **languages** | **jit level** | **graph compile** | **ms/step** | **img/s** | **f-score** | **recipe** | **download** | +|----------------|-----------|----------------|--------------|:-----------------:|-------------|:-----------------:|-------------|------------|-------------|-----------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------| +| DBNet | ResNet-50 | 8 | 10 | Chinese + English | O2| 107.91 s | 312.48 | 256 | 83.41% | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141-912f0a90.mindir) | +| DBNet++ | ResNet-50 | 4 | 32 | Chinese + English | O2| 182.94 s | 1230.76 | 104| 84.30% | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9-b3f52398.mindir) |
> The input_shape for exported DBNet MindIR and DBNet++ MindIR in the links are `(1,3,736,1280)` and `(1,3,1152,2048)`, respectively. @@ -88,21 +88,21 @@ Performance tested on ascend 910 with graph mode DBNet and DBNet++ were trained on the ICDAR2015, MSRA-TD500, SCUT-CTW1500, Total-Text, and MLT2017 datasets. In addition, we conducted pre-training on the SynthText dataset and provided a URL to download pretrained weights. All training results are as follows:
- Performance tested on ascend 910 with graph mode + Experiments are tested on ascend 910 with mindspore 2.3.1 graph mode ### ICDAR2015
- | **Model** | **Cards** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |---------------------|-----|---------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| - | DBNet | 1 | MobileNetV3 | ImageNet | 76.31% | 78.27% | 77.28% | 10 | 100.00 ms/step | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539-f14c6a13.mindir) | - | DBNet | 8 | MobileNetV3 | ImageNet | 76.22% | 77.98% | 77.09% | 8 | 66.64 ms/step | [yaml](db_mobilenetv3_icdar15_8p.yaml) | Coming soon | - | DBNet | 1 | ResNet-18 | ImageNet | 80.12% | 83.41% | 81.73% | 20 | 185.19 ms/step | [yaml](db_r18_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa-cf46eb8b.mindir) | - | DBNet | 1 | ResNet-50 | ImageNet | 83.53% | 86.62% | 85.05% | 10 | 132.98 ms/step | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24-fbf95c82.mindir) | - | DBNet | 8 | ResNet-50 | ImageNet | 82.62% | 88.54% | 85.48% | 10 | 183.92 ms/step | [yaml](db_r50_icdar15_8p.yaml) | Coming soon | - | | | | | | | | | | | | - | DBNet++ | 1 | ResNet-50 | SynthText | 86.81% | 86.85% | 86.86% | 32 | 409.21 ms/step | [yaml](dbpp_r50_icdar15_910.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2-e61a9c37.mindir) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | img/s | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|-----------|----------------|--------------|----------------|------------|---------------|-------------|-------------|-----------------------------------------|----------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| + | DBNet | MobileNetV3 | 1 | 10 | ImageNet | 100 | 100 | 76.31% | 78.27% | 77.28% | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539-f14c6a13.mindir) | + | DBNet | MobileNetV3 | 8 | 8 | ImageNet | 66.64 | 960 | 76.22% | 77.98% | 77.09% | [yaml](db_mobilenetv3_icdar15_8p.yaml) | Coming soon | + | DBNet | ResNet-18 | 1 | 20 | ImageNet | 185.19 | 108 | 80.12% | 83.41% | 81.73% | [yaml](db_r18_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa-cf46eb8b.mindir) | + | DBNet | ResNet-50 | 1 | 10 | ImageNet | 132.98 | 75.2 | 83.53% | 86.62% | 85.05% | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24-fbf95c82.mindir) | + | DBNet | ResNet-50 | 8 | 10 | ImageNet | 183.92 | 435 | 82.62% | 88.54% | 85.48% | [yaml](db_r50_icdar15_8p.yaml) | Coming soon | + | | | | | | | | | | | | + | DBNet++ | ResNet-50 | 1 | 32 | SynthText | 409.21 | 78,.2| 86.81% | 86.85% | 86.86% | [yaml](dbpp_r50_icdar15_910.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2-e61a9c37.mindir) |
> The input_shape for exported DBNet MindIR and DBNet++ MindIR in the links are `(1,3,736,1280)` and `(1,3,1152,2048)`, respectively. @@ -111,10 +111,10 @@ DBNet and DBNet++ were trained on the ICDAR2015, MSRA-TD500, SCUT-CTW1500, Total
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|---------------------------|-------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 79.90% | 88.07% | 83.78% | 20 | 164.34 ms/step | [yaml](db_r18_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_td500-b5abff68.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 84.02% | 87.48% | 85.71% | 20 | 280.90 ms/step | [yaml](db_r50_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_td500-0d12b5e8.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|-------------|-------------|------------|-------------|------------------|---------------------------|-------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 163.34 | 121.7 | 79.90% | 88.07% | 83.78% | [yaml](db_r18_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_td500-b5abff68.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 280.90 | 71.2| 84.02% | 87.48% | 85.71% | [yaml](db_r50_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_td500-0d12b5e8.ckpt) |
> MSRA-TD500 dataset has 300 training images and 200 testing images, reference paper [Real-time Scene Text Detection with Differentiable Binarization](https://arxiv.org/abs/1911.08947), we trained using an extra 400 traning images from HUST-TR400. You can down all [dataset](https://paddleocr.bj.bcebos.com/dataset/TD_TR.tar) for training. @@ -123,57 +123,57 @@ DBNet and DBNet++ were trained on the ICDAR2015, MSRA-TD500, SCUT-CTW1500, Total
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-----------------------------|---------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 85.68% | 85.33% | 85.50% | 20 | 163.80 ms/step | [yaml](db_r18_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_ctw1500-0864b040.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 87.83% | 84.71% | 86.25% | 20 | 180.11 ms/step | [yaml](db_r50_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ctw1500-f637e3d3.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|-------------|------------|-------------|-------------|-----------------------------|---------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 163.80 | 122.1 | 85.68% | 85.33% | 85.50% | [yaml](db_r18_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_ctw1500-0864b040.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 180.11 | 71.4| 87.83% | 84.71% | 86.25% | [yaml](db_r50_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ctw1500-f637e3d3.ckpt) |
### Total-Text
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------|-----------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 83.66% | 87.61% | 85.59% | 20 | 206.40 ms/step | [yaml](db_r18_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_totaltext-fb456ff4.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 84.79% | 87.07% | 85.91% | 20 | 289.44 ms/step | [yaml](db_r50_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_totaltext-76d6f421.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|-------------|------------|-------------|-------------|-------------------------------|-----------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 206.40 | 96.9 | 83.66% | 87.61% | 85.59% | [yaml](db_r18_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_totaltext-fb456ff4.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 289.44 | 69.1| 84.79% | 87.07% | 85.91% | [yaml](db_r50_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_totaltext-76d6f421.ckpt) |
### MLT2017
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-----------------------------|---------------------------------------------------------------------------------------------------| - | DBNet | 8p | ResNet-18 | SynthText | 73.62% | 83.93% | 78.44% | 20 | 464.00 ms/step | [yaml](db_r18_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_mlt2017-5af33809.ckpt) | - | DBNet | 8p | ResNet-50 | SynthText | 76.04% | 84.51% | 80.05% | 20 | 523.6 ms/step | [yaml](db_r50_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_mlt2017-3bd6e569.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|--------------|------------|-------------|-------------|-----------------------------|---------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 8 | 20 | SynthText | 464.00 | 344.8 | 73.62% | 83.93% | 78.44% | [yaml](db_r18_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_mlt2017-5af33809.ckpt) | + | DBNet | ResNet-50 | 8 | 20 | SynthText | 523.60 | 305.6| 76.04% | 84.51% | 80.05% | [yaml](db_r50_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_mlt2017-3bd6e569.ckpt) |
### SynthText
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Train Loss**| **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-------------------|------------|--------------|----------------|-------------|----------------|----------------|-------------|--------------| - | DBNet | 1p | ResNet-18 | ImageNet | 2.41 | 16 | 131.83 ms/step | [yaml](db_r18_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_synthtext-251ef3dd.ckpt) | - | DBNet | 1p | ResNet-50 | ImageNet | 2.25 | 16 | 195.07 ms/step | [yaml](db_r50_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_synthtext-40655acb.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **train loss** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|----------------|------------|----------------|-------------------------------|-----------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 16 | ImageNet | 131.83 | 121.37 | 2.41 | [yaml](db_r18_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_synthtext-251ef3dd.ckpt) | + | DBNet | ResNet-50 | 1 | 16 | ImageNet | 195.07 | 82.02| 2.25 | [yaml](db_r50_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_synthtext-40655acb.ckpt) |
- Performance tested on ascend 910* with graph mode + Experiments are tested on ascend 910* with mindspore 2.3.1 graph mode ### ICDAR2015
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |---------------------|-----------------|---------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------------|------------------------------------------------------------------------------------------------------------| - | DBNet | 1p | MobileNetV3 | ImageNet | 74.68% | 79.38% | 76.95% | 10 | 65.69 ms/step | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-e72f9b8b-910v2.ckpt) | - | DBNet | 8p | MobileNetV3 | ImageNet | 76.27% | 76.06% | 76.17% | 8 | 54.46 ms/step | [yaml](db_mobilenetv3_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-7e89e1df-910v2.ckpt) | - | DBNet | 1p | ResNet-50 | ImageNet | 84.50% | 85.36% | 84.93% | 10 | 155.62 ms/step | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-48153c3b-910v2.ckpt) | - | DBNet | 8p | ResNet-50 | ImageNet | 81.15% | 87.63% | 84.26% | 10 | 159.22 ms/step | [yaml](db_r50_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-e10bad35-910v2.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|-------|----------------|--------------|----------------|------------|-----------|------------|-------------|-------------|----------------------------------------|------------------------------------------------------------------------------------------------------------| + | DBNet | MobileNetV3 | 1 | 10 | ImageNet | 65.69 | 152.23 | 74.68% | 79.38% | 76.95% | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-e72f9b8b-910v2.ckpt) | + | DBNet | MobileNetV3 | 8 | 8 | ImageNet | 54.46 | 1175.12 | 76.27% | 76.06% | 76.17% | [yaml](db_mobilenetv3_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-7e89e1df-910v2.ckpt) | + | DBNet | ResNet-50 | 1 | 10 | ImageNet | 155.62 | 64.25 | 84.50% | 85.36% | 84.93% | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-48153c3b-910v2.ckpt) | + | DBNet | ResNet-50 | 8 | 10 | ImageNet | 159.22 | 502.4 | 81.15% | 87.63% | 84.26% | [yaml](db_r50_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-e10bad35-910v2.ckpt) |
diff --git a/configs/det/dbnet/README_CN.md b/configs/det/dbnet/README_CN.md index a48549d04..396e42b8f 100644 --- a/configs/det/dbnet/README_CN.md +++ b/configs/det/dbnet/README_CN.md @@ -54,13 +54,13 @@ DBNet++在检测不同尺寸的文本方面表现更好，尤其是对于尺寸 这些模型在12个公开数据集上训练，包括CTW，LSVT，RCTW-17，TextOCR等，其中训练集包含153511张图片，验证集包含9786张图片。
从上述数据集中手动选择598张未被训练集和验证集使用的图片构成测试集。 -在采用图模式的ascend 910上测试性能 +在采用图模式的ascend 910上实验结果，mindspore版本为2.3.1
-| **Model** | **Cards** | **Backbone** | **Languages** | **F-score** | **Batch Size** | **graph compile** | **jit level** | **Step Time** | **Recipe** | **Download** | -|-----------|---------|--------------|-------------------|:---------------------------:|----------------|:-----------------:|--|------------------|----------------|----------| -| DBNet | 8 | ResNet-50 | Chinese + English | 83.41% | 10 | 107.91 s | O2| 312.48 ms/step | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141-912f0a90.mindir) | -| DBNet++ | 4 | ResNet-50 | Chinese + English | 84.30% | 32 | 182.94 s | O2| 1230.76 ms/step | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9-b3f52398.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **languages** | **jit level** | **graph compile** | **ms/step** | **img/s** | **f-score** | **recipe** | **download** | +|----------------|-----------|----------------|--------------|:-----------------:|-------------|:-----------------:|-------------|------------|-------------|-----------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------| +| DBNet | ResNet-50 | 8 | 10 | Chinese + English | O2| 107.91 s | 312.48 | 256 | 83.41% | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ch_en_general-a5dbb141-912f0a90.mindir) | +| DBNet++ | ResNet-50 | 4 | 32 | Chinese + English | O2| 182.94 s | 1230.76 | 104| 84.30% | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9.ckpt) | [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_ch_en_general-884ba5b9-b3f52398.mindir) |
> 链接中模型DBNet的MindIR导出时的输入Shape为`(1,3,736,1280)`，模型DBNet++的MindIR导出时的输入Shape为`(1,3,1152,2048)`。 @@ -71,21 +71,21 @@ DBNet++在检测不同尺寸的文本方面表现更好，尤其是对于尺寸 DBNet和DBNet++在ICDAR2015，MSRA-TD500，SCUT-CTW1500，Total-Text和MLT2017数据集上训练。另外，我们在SynthText数据集上进行了预训练，并提供预训练权重下载链接。所有训练结果如下：
- 在采用图模式的ascend 910上测试性能 + 在采用图模式的ascend 910上实验结果，mindspore版本为2.3.1 ### ICDAR2015
- | **Model** | **Cards** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |---------------------|-----|---------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| - | DBNet | 1 | MobileNetV3 | ImageNet | 76.31% | 78.27% | 77.28% | 10 | 100.00 ms/step | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539-f14c6a13.mindir) | - | DBNet | 8 | MobileNetV3 | ImageNet | 76.22% | 77.98% | 77.09% | 8 | 66.64 ms/step | [yaml](db_mobilenetv3_icdar15_8p.yaml) | Coming soon | - | DBNet | 1 | ResNet-18 | ImageNet | 80.12% | 83.41% | 81.73% | 20 | 185.19 ms/step | [yaml](db_r18_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa-cf46eb8b.mindir) | - | DBNet | 1 | ResNet-50 | ImageNet | 83.53% | 86.62% | 85.05% | 10 | 132.98 ms/step | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24-fbf95c82.mindir) | - | DBNet | 8 | ResNet-50 | ImageNet | 82.62% | 88.54% | 85.48% | 10 | 183.92 ms/step | [yaml](db_r50_icdar15_8p.yaml) | Coming soon | - | | | | | | | | | | | | - | DBNet++ | 1 | ResNet-50 | SynthText | 86.81% | 86.85% | 86.86% | 32 | 409.21 ms/step | [yaml](dbpp_r50_icdar15_910.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2-e61a9c37.mindir) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | img/s | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|-----------|----------------|--------------|----------------|------------|---------------|-------------|-------------|-----------------------------------------|----------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| + | DBNet | MobileNetV3 | 1 | 10 | ImageNet | 100 | 100 | 76.31% | 78.27% | 77.28% | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-62c44539-f14c6a13.mindir) | + | DBNet | MobileNetV3 | 8 | 8 | ImageNet | 66.64 | 960 | 76.22% | 77.98% | 77.09% | [yaml](db_mobilenetv3_icdar15_8p.yaml) | Coming soon | + | DBNet | ResNet-18 | 1 | 20 | ImageNet | 185.19 | 108 | 80.12% | 83.41% | 81.73% | [yaml](db_r18_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18-0c0c4cfa-cf46eb8b.mindir) | + | DBNet | ResNet-50 | 1 | 10 | ImageNet | 132.98 | 75.2 | 83.53% | 86.62% | 85.05% | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50-c3a4aa24-fbf95c82.mindir) | + | DBNet | ResNet-50 | 8 | 10 | ImageNet | 183.92 | 435 | 82.62% | 88.54% | 85.48% | [yaml](db_r50_icdar15_8p.yaml) | Coming soon | + | | | | | | | | | | | | + | DBNet++ | ResNet-50 | 1 | 32 | SynthText | 409.21 | 78,.2| 86.81% | 86.85% | 86.86% | [yaml](dbpp_r50_icdar15_910.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnetpp_resnet50_910-35dc71f2-e61a9c37.mindir) |
> 链接中模型DBNet的MindIR导出时的输入Shape为`(1,3,736,1280)`，模型DBNet++的MindIR导出时的输入Shape为`(1,3,1152,2048)`。 @@ -94,10 +94,10 @@ DBNet和DBNet++在ICDAR2015，MSRA-TD500，SCUT-CTW1500，Total-Text和MLT2017
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|---------------------------|-------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 79.90% | 88.07% | 83.78% | 20 | 164.34 ms/step | [yaml](db_r18_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_td500-b5abff68.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 84.02% | 87.48% | 85.71% | 20 | 280.90 ms/step | [yaml](db_r50_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_td500-0d12b5e8.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|-------------|-------------|------------|-------------|------------------|---------------------------|-------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 163.34 | 121.7 | 79.90% | 88.07% | 83.78% | [yaml](db_r18_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_td500-b5abff68.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 280.90 | 71.2| 84.02% | 87.48% | 85.71% | [yaml](db_r50_td500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_td500-0d12b5e8.ckpt) |
> MSRA-TD500数据集有300训练集图片和200测试集图片，参考论文[Real-time Scene Text Detection with Differentiable Binarization](https://arxiv.org/abs/1911.08947)，我们训练此权重额外使用了来自HUST-TR400数据集的400训练集图片。可以在此下载全部[数据集](https://paddleocr.bj.bcebos.com/dataset/TD_TR.tar)用于训练。 @@ -106,57 +106,57 @@ DBNet和DBNet++在ICDAR2015，MSRA-TD500，SCUT-CTW1500，Total-Text和MLT2017
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-----------------------------|---------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 85.68% | 85.33% | 85.50% | 20 | 163.80 ms/step | [yaml](db_r18_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_ctw1500-0864b040.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 87.83% | 84.71% | 86.25% | 20 | 180.11 ms/step | [yaml](db_r50_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ctw1500-f637e3d3.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|-------------|------------|-------------|-------------|-----------------------------|---------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 163.80 | 122.1 | 85.68% | 85.33% | 85.50% | [yaml](db_r18_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_ctw1500-0864b040.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 180.11 | 71.4| 87.83% | 84.71% | 86.25% | [yaml](db_r50_ctw1500.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_ctw1500-f637e3d3.ckpt) |
### Total-Text
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------|-----------------------------------------------------------------------------------------------------| - | DBNet | 1p | ResNet-18 | SynthText | 83.66% | 87.61% | 85.59% | 20 | 206.40 ms/step | [yaml](db_r18_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_totaltext-fb456ff4.ckpt) | - | DBNet | 1p | ResNet-50 | SynthText | 84.79% | 87.07% | 85.91% | 20 | 289.44 ms/step | [yaml](db_r50_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_totaltext-76d6f421.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|-------------|------------|-------------|-------------|-------------------------------|-----------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 20 | SynthText | 206.40 | 96.9 | 83.66% | 87.61% | 85.59% | [yaml](db_r18_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_totaltext-fb456ff4.ckpt) | + | DBNet | ResNet-50 | 1 | 20 | SynthText | 289.44 | 69.1| 84.79% | 87.07% | 85.91% | [yaml](db_r50_totaltext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_totaltext-76d6f421.ckpt) |
### MLT2017
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-----------|------------|--------------|----------------|------------|---------------|-------------|----------------|----------------|-----------------------------|---------------------------------------------------------------------------------------------------| - | DBNet | 8p | ResNet-18 | SynthText | 73.62% | 83.93% | 78.44% | 20 | 464.00 ms/step | [yaml](db_r18_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_mlt2017-5af33809.ckpt) | - | DBNet | 8p | ResNet-50 | SynthText | 76.04% | 84.51% | 80.05% | 20 | 523.6 ms/step | [yaml](db_r50_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_mlt2017-3bd6e569.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|------------|--------------|------------|-------------|-------------|-----------------------------|---------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 8 | 20 | SynthText | 464.00 | 344.8 | 73.62% | 83.93% | 78.44% | [yaml](db_r18_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_mlt2017-5af33809.ckpt) | + | DBNet | ResNet-50 | 8 | 20 | SynthText | 523.60 | 305.6| 76.04% | 84.51% | 80.05% | [yaml](db_r50_mlt2017.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_mlt2017-3bd6e569.ckpt) |
### SynthText
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Train Loss**| **Batch Size** | **Step Time** | **Recipe** | **Download** | - |-------------------|------------|--------------|----------------|-------------|----------------|----------------|-------------|--------------| - | DBNet | 1p | ResNet-18 | ImageNet | 2.41 | 16 | 131.83 ms/step | [yaml](db_r18_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_synthtext-251ef3dd.ckpt) | - | DBNet | 1p | ResNet-50 | ImageNet | 2.25 | 16 | 195.07 ms/step | [yaml](db_r50_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_synthtext-40655acb.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **train loss** | **recipe** | **download** | + |----------------|---------|----------------|--------------|----------------|----------------|------------|----------------|-------------------------------|-----------------------------------------------------------------------------------------------------| + | DBNet | ResNet-18 | 1 | 16 | ImageNet | 131.83 | 121.37 | 2.41 | [yaml](db_r18_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet18_synthtext-251ef3dd.ckpt) | + | DBNet | ResNet-50 | 1 | 16 | ImageNet | 195.07 | 82.02| 2.25 | [yaml](db_r50_synthtext.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/dbnet/dbnet_resnet50_synthtext-40655acb.ckpt) |
- 在采用图模式的ascend 910*上测试性能 + 在采用图模式的ascend 910*上实验结果，mindspore版本为2.3.1 ### ICDAR2015
- | **Model** | **Device Card** | **Backbone** | **Pretrained** | **Recall** | **Precision** | **F-score** | **Batch Size** | **Step Time** | **Recipe** | **Download** | - |---------------------|-----------------|---------------|----------------|------------|---------------|-------------|----------------|----------------|-------------------------------------|------------------------------------------------------------------------------------------------------------| - | DBNet | 1p | MobileNetV3 | ImageNet | 74.68% | 79.38% | 76.95% | 10 | 65.69 ms/step | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-e72f9b8b-910v2.ckpt) | - | DBNet | 8p | MobileNetV3 | ImageNet | 76.27% | 76.06% | 76.17% | 8 | 54.46 ms/step | [yaml](db_mobilenetv3_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-7e89e1df-910v2.ckpt) | - | DBNet | 1p | ResNet-50 | ImageNet | 84.50% | 85.36% | 84.93% | 10 | 155.62 ms/step | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-48153c3b-910v2.ckpt) | - | DBNet | 8p | ResNet-50 | ImageNet | 81.15% | 87.63% | 84.26% | 10 | 159.22 ms/step | [yaml](db_r50_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-e10bad35-910v2.ckpt) | + | **model name** | **backbone** | **cards** | **batch size** | **pretrained** | **ms/step** | **img/s** | **recall** | **precision** | **f-score** | **recipe** | **download** | + |----------------|-------|----------------|--------------|----------------|------------|-----------|------------|-------------|-------------|----------------------------------------|------------------------------------------------------------------------------------------------------------| + | DBNet | MobileNetV3 | 1 | 10 | ImageNet | 65.69 | 152.23 | 74.68% | 79.38% | 76.95% | [yaml](db_mobilenetv3_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-e72f9b8b-910v2.ckpt) | + | DBNet | MobileNetV3 | 8 | 8 | ImageNet | 54.46 | 1175.12 | 76.27% | 76.06% | 76.17% | [yaml](db_mobilenetv3_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_mobilenetv3-7e89e1df-910v2.ckpt) | + | DBNet | ResNet-50 | 1 | 10 | ImageNet | 155.62 | 64.25 | 84.50% | 85.36% | 84.93% | [yaml](db_r50_icdar15.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-48153c3b-910v2.ckpt) | + | DBNet | ResNet-50 | 8 | 10 | ImageNet | 159.22 | 502.4 | 81.15% | 87.63% | 84.26% | [yaml](db_r50_icdar15_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/dbnet/dbnet_resnet50-e10bad35-910v2.ckpt) |
diff --git a/configs/rec/crnn/README.md b/configs/rec/crnn/README.md index e7f644218..aea9ea03e 100644 --- a/configs/rec/crnn/README.md +++ b/configs/rec/crnn/README.md @@ -43,32 +43,32 @@ Table Format: According to our experiments, the training (following the steps in [Model Training](#32-model-training)) performance and evaluation (following the steps in [Model Evaluation](#33-model-evaluation)) accuracy are as follows: -#### Performance tested on ascend 910 with graph mode +#### Experiments are tested on ascend 910 with mindspore 2.3.1 graph mode
-| **Model** | **Cards** | **Backbone** | **Train Dataset** | **Model Params** | **Batch size** | **graph compile** | **jit level** | **Step Time** | **FPS** | **Avg Eval Accuracy** | **Recipe** | **Download** | -| :-----: |:---------:| :-----: | :-----: | :-----: | :-----: |:-----------------:|:-------:|:-------------:| :-----: | :-----: | :-----: | :-----: | -| CRNN | 8 | VGG7 | MJ+ST | 8.72 M | 16 | 67.18 s | O2| 22.06 ms/step | 5802.71 | 82.03% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c-573dbd61.mindir) | -| CRNN | 8 | ResNet34_vd | MJ+ST | 24.48 M | 64 | 201.54 s | O2| 76.48 ms/step | 6694.84 | 84.45% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07-eb10a0c9.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **train dataset** | **model params** | **jit level** | **graph compile** | **ms/step** | **img/s** | **avg eval accuracy** | **recipe** |**download** | +|:--------------:|:---------:|:--------------:| :-----: |:-----------------:|:----------------:|:---------------------:|:-------:|:-----------:|:---------:|:---------------------:|:-------------------------:|:----------------------------------------------------:| +| CRNN | VGG7 | 8 | 16 | MJ+ST | 8.72 M | O2| 67.18 s | 22.06 | 5802.71 | 82.03% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c-573dbd61.mindir) | +| CRNN | ResNet34_vd | 8 | 64 | MJ+ST | 24.48 M | O2| 201.54 s | 76.48 | 6694.84 | 84.45% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07-eb10a0c9.mindir) |
- Detailed accuracy results for each benchmark dataset (IC03, IC13, IC15, IIIT, SVT, SVTP, CUTE):
- | **Model** | **Backbone** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **Average** | - | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | - | CRNN | VGG7 | 94.53% | 94.00% | 92.18% | 90.74% | 71.95% | 66.06% | 84.10% | 83.93% | 73.33% | 69.44% | 82.03% | - | CRNN | ResNet34_vd | 94.42% | 94.23% | 93.35% | 92.02% | 75.92% | 70.15% | 87.73% | 86.40% | 76.28% | 73.96% | 84.45% | + | **model name** | **backbone** | **cards** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **average** | + |:--------------:| :------: |:------------:|:------------:| :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: |:-----------:| + | CRNN | VGG7 |1| 94.53% | 94.00% | 92.18% | 90.74% | 71.95% | 66.06% | 84.10% | 83.93% | 73.33% | 69.44% | 82.03% | + | CRNN | ResNet34_vd |1| 94.42% | 94.23% | 93.35% | 92.02% | 75.92% | 70.15% | 87.73% | 86.40% | 76.28% | 73.96% | 84.45% |
-#### Performance tested on ascend 910* with graph mode +#### Experiments are tested on ascend 910* with mindspore 2.3.1 graph mode
-| **Model** | **Cards** | **Backbone** | **Train Dataset** | **Model Params** | **Batch size** | **graph compile** | **jit level** | **Step Time** | **FPS** | **Avg Eval Accuracy** | **Recipe** | **Download** | -| :-----: |:---------:| :-----: | :-----: | :-----: | :-----: |:-----------------:|:----------------------------:|:-------------:|:------------------------:|:---------------------:| :-----: | :-----: | -| CRNN | 8 | VGG7 | MJ+ST | 8.72 M | 16 | 94.36 s | O2| 14.76 ms/step | 8672.09 | 81.31% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/crnn/crnn_vgg7-6faf1b2d-910v2.ckpt)| +| **model name** | **backbone** | **cards** | **batch size** | **train dataset** | **model params** | **jit level** | **graph compile** | **ms/step** | **img/s** | **avg eval accuracy** |**recipe** | **download** | +|:--------------:|:---------:|:--------------:|:------------:|:-----------------:|:----------------:|:---------------------:|:----------------------------:|:-----------:|:---------:|:---------------------:|:------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------:| +| CRNN | VGG7 | 8 | 16 | MJ+ST | 8.72 M | O2 | 94.36 s | 14.76 | 8672.09 | 81.31% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/crnn/crnn_vgg7-6faf1b2d-910v2.ckpt) |
@@ -77,12 +77,14 @@ According to our experiments, the training (following the steps in [Model Traini The inference performance is tested on Mindspore Lite, please take a look at [Mindpore Lite Inference](#6-mindspore-lite-inference) for more details. +Experiments are tested on ascend 310P with mindspore lite 2.3.1 graph mode +
-| Device | Env | Model | Backbone | Params | Test Dataset | Batch size | Graph infer 1P (FPS) | -| :----: | :----: | :----: | :----: | :----: | :----: | :----: | :----: | -| Ascend310P | Lite2.0 | CRNN | ResNet34_vd | 24.48 M | IC15 | 1 | 361.09 | -| Ascend310P | Lite2.0 | CRNN | ResNet34_vd | 24.48 M | SVT | 1 | 274.67 | +| model name | backbone | batch size | params | test dataset | img/s | +|:----------:|:----------:|:-----------:|:-------:|:------------:|:------:| +| CRNN | ResNet34_vd | 1 | 24.48 M | IC15 | 361.09 | +| CRNN | ResNet34_vd | 1 | 24.48 M | SVT | 274.67 |
@@ -397,9 +399,9 @@ After training, evaluation results on the benchmark test set are as follows, whe
-| **Model** | **Language** | **Context** |**Backbone** | **Scene** | **Web** | **Document** | **Train T.** | **FPS** | **Recipe** | **Download** | -| :-----: | :-----: | :--------: | :--------: | :--------: | :--------: | :--------: | :---------: | :--------: | :---------: | :-----------: | -| CRNN | Chinese | D910x4-MS1.10-G | ResNet34_vd | 60.45% | 65.95% | 97.68% | 647 s/epoch | 1180 | [crnn_resnet34_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c-105bccb2.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **language** | **jit level** | **graph compile** | **ms/step** | **img/s** | **scene** | **web** | **document** | **recipe** | **download** | +|:--------------:|:------------:|:--------------:|:-----------------:|:------------:|:---------:|:-----------------:|:---------:|:-------:|:------------:|:-----------:|:---------:|:------------------------------------------------------------------------------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| CRNN | ResNet34_vd | 4| 256| Chinese | O2 | 203.48 s | 38.01 | 1180 | 60.45% | 65.95% | 97.68% | [crnn_resnet34_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c-105bccb2.mindir) |
**Notes:** diff --git a/configs/rec/crnn/README_CN.md b/configs/rec/crnn/README_CN.md index 6dcfb4f3e..b282c6b01 100644 --- a/configs/rec/crnn/README_CN.md +++ b/configs/rec/crnn/README_CN.md @@ -44,35 +44,35 @@ Table Format: 根据我们的实验，训练（[模型训练](#32-模型训练)）性能和精度评估（[模型评估](#33-模型评估)）结果如下： -#### 在采用图模式的ascend 910上测试性能 +#### 在采用图模式的ascend 910上实验结果，mindspore版本为2.3.1
-| **Model** | **Cards** | **Backbone** | **Train Dataset** | **Model Params** | **Batch size** | **graph compile** | **jit level** | **Step Time** | **FPS** | **Avg Eval Accuracy** | **Recipe** | **Download** | -| :-----: |:---------:| :-----: | :-----: | :-----: | :-----: |:-----------------:|:-------:|:-------------:| :-----: | :-----: | :-----: | :-----: | -| CRNN | 8 | VGG7 | MJ+ST | 8.72 M | 16 | 67.18 s | O2| 22.06 ms/step | 5802.71 | 82.03% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c-573dbd61.mindir) | -| CRNN | 8 | ResNet34_vd | MJ+ST | 24.48 M | 64 | 201.54 s | O2| 76.48 ms/step | 6694.84 | 84.45% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07-eb10a0c9.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **train dataset** | **model params** | **jit level** | **graph compile** | **ms/step** | **img/s** | **avg eval accuracy** | **recipe** |**download** | +|:--------------:|:---------:|:--------------:| :-----: |:-----------------:|:----------------:|:---------------------:|:-------:|:-----------:|:---------:|:---------------------:|:-------------------------:|:----------------------------------------------------:| +| CRNN | VGG7 | 8 | 16 | MJ+ST | 8.72 M | O2| 67.18 s | 22.06 | 5802.71 | 82.03% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_vgg7-ea7e996c-573dbd61.mindir) | +| CRNN | ResNet34_vd | 8 | 64 | MJ+ST | 24.48 M | O2| 201.54 s | 76.48 | 6694.84 | 84.45% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34-83f37f07-eb10a0c9.mindir) |
- 在各个基准数据集（IC03，IC13，IC15，IIIT，SVT，SVTP，CUTE）上的准确率：
- | **Model** | **Backbone** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **Average** | - | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | - | CRNN | VGG7 | 94.53% | 94.00% | 92.18% | 90.74% | 71.95% | 66.06% | 84.10% | 83.93% | 73.33% | 69.44% | 82.03% | - | CRNN | ResNet34_vd | 94.42% | 94.23% | 93.35% | 92.02% | 75.92% | 70.15% | 87.73% | 86.40% | 76.28% | 73.96% | 84.45% | + | **model name** | **backbone** | **cards** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **average** | + |:--------------:| :------: |:------------:|:------------:| :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: |:-----------:| + | CRNN | VGG7 |1| 94.53% | 94.00% | 92.18% | 90.74% | 71.95% | 66.06% | 84.10% | 83.93% | 73.33% | 69.44% | 82.03% | + | CRNN | ResNet34_vd |1| 94.42% | 94.23% | 93.35% | 92.02% | 75.92% | 70.15% | 87.73% | 86.40% | 76.28% | 73.96% | 84.45% |
-#### 在采用图模式的ascend 910*上测试性能 +#### 在采用图模式的ascend 910*上实验结果，mindspore版本为2.3.1
-| **Model** | **Cards** | **Backbone** | **Train Dataset** | **Model Params** | **Batch size** | **graph compile** | **jit level** | **Step Time** | **FPS** | **Avg Eval Accuracy** | **Recipe** | **Download** | -| :-----: |:---------:| :-----: | :-----: | :-----: | :-----: |:-----------------:|:----------------------------:|:-------------:|:------------------------:|:---------------------:| :-----: | :-----: | -| CRNN | 8 | VGG7 | MJ+ST | 8.72 M | 16 | 94.36 s | O2| 14.76 ms/step | 8672.09 | 81.31% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/crnn/crnn_vgg7-6faf1b2d-910v2.ckpt)| +| **model name** | **backbone** | **cards** | **batch size** | **train dataset** | **model params** | **jit level** | **graph compile** | **ms/step** | **img/s** | **avg eval accuracy** |**recipe** | **download** | +|:--------------:|:---------:|:--------------:|:------------:|:-----------------:|:----------------:|:---------------------:|:----------------------------:|:-----------:|:---------:|:---------------------:|:------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------:| +| CRNN | VGG7 | 8 | 16 | MJ+ST | 8.72 M | O2 | 94.36 s | 14.76 | 8672.09 | 81.31% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_vgg7.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/crnn/crnn_vgg7-6faf1b2d-910v2.ckpt) |
@@ -80,12 +80,14 @@ Table Format: 推理端的性能测试主要是基于Mindspore Lite，详细的操作介绍可参考 [Mindspore Lite推理](#6-mindspore-lite-推理)。 +在采用图模式的ascend 310P上实验结果，mindspore lite版本为2.3.1 +
-| 设备 | 编译环境 | 模型 | 骨干网络 | 参数量 | 测试集 | 批量大小 | 图模式单卡推理 (FPS) | -| :----: | :----: | :----: | :----: | :----: | :----: | :----: | :----: | -| Ascend310P | Lite2.0 | CRNN | ResNet34_vd | 24.48 M | IC15 | 1 | 361.09 | -| Ascend310P | Lite2.0 | CRNN | ResNet34_vd | 24.48 M | SVT | 1 | 274.67 | +| model name | backbone | batch size | params | test dataset | img/s | +|:----------:|:----------:|:-----------:|:-------:|:------------:|:------:| +| CRNN | ResNet34_vd | 1 | 24.48 M | IC15 | 361.09 | +| CRNN | ResNet34_vd | 1 | 24.48 M | SVT | 274.67 |
@@ -399,9 +401,9 @@ mpirun --allow-run-as-root -n 4 python tools/train.py --config configs/rec/crnn/
-| **模型** | **语种** | **环境配置** | **骨干网络** | **街景类** | **网页类** | **文档类** | **训练时间** | **FPS** | **配置文件** | **模型权重下载** | -| :-----: | :-----: | :-------: |:--------: | :--------: | :--------: | :--------: | :---------: |:--------: | :---------: | :-----------: | -| CRNN | 中文 | D910x4-MS1.10-G | ResNet34_vd | 60.45% | 65.95% | 97.68% | 647 s/epoch | 1180 | [crnn_resnet34_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34_ch.yaml) |[ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c-105bccb2.mindir) | +| **model name** | **backbone** | **cards** | **batch size** | **language** | **jit level** | **graph compile** | **ms/step** | **img/s** | **scene** | **web** | **document** | **recipe** | **download** | +|:--------------:|:------------:|:--------------:|:-----------------:|:------------:|:---------:|:-----------------:|:---------:|:-------:|:------------:|:-----------:|:---------:|:------------------------------------------------------------------------------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| CRNN | ResNet34_vd | 4| 256| Chinese | O2 | 203.48 s | 38.01 | 1180 | 60.45% | 65.95% | 97.68% | [crnn_resnet34_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/crnn/crnn_resnet34_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/crnn/crnn_resnet34_ch-7a342e3c-105bccb2.mindir) |
**注释：** diff --git a/configs/rec/svtr/README.md b/configs/rec/svtr/README.md index 7d5df6c9e..6a954b844 100644 --- a/configs/rec/svtr/README.md +++ b/configs/rec/svtr/README.md @@ -41,23 +41,23 @@ Table Format: According to our experiments, the evaluation results on public benchmark datasets (IC03, IC13, IC15, IIIT, SVT, SVTP, CUTE) is as follow: -#### Performance tested on ascend 910 with graph mode +#### Experiments are tested on ascend 910 with mindspore 2.3.1 graph mode
-| **Model** | **Cards** | **Avg Accuracy** | **Batch size** | **graph compile** |**jit level** | **Step Time** | **FPS** | **Recipe** | **Download** | -| :-----: |:---------:| :--------------: |:--------------:|:-----------------:|:-----------------:|:-------------:| :--------: | :--------: |:----------: | -| SVTR-Tiny | 4 | 90.23% | 512 | 226.86 s |O2| 49.38 ms/step | 4560 | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3-86ece8c8.mindir) | -| SVTR-Tiny-8P | 8 | 90.32% | 512 | 226.86 s |O2| 55.16 ms/step | 9840 | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6.ckpt) \| [mindir](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6-255191ef.mindir) | +| **model name** | **cards** | **batch size** |**jit level** | **graph compile** | **ms/step** | **img/s** | **avg accuracy** | **recipe** | **download** | +|:--------------:|:---------:|:--------------:|:----------------:|:-----------------:|:-----------------:|:-------------:|:---------:|:---------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| SVTR-Tiny | 4 | 512 |O2| 226.86 s | 49.38 ms/step | 4560 | 90.23% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3-86ece8c8.mindir) | +| SVTR-Tiny-8P | 8 | 512 |O2| 226.86 s | 55.16 ms/step | 9840 | 90.32% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6.ckpt) \| [mindir](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6-255191ef.mindir) |
Detailed accuracy results for each benchmark dataset
-| **Model** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **Average** | -| :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | -| SVTR-Tiny | 95.70% | 95.50% | 95.33% | 93.99% | 83.60% | 79.83% | 94.70% | 91.96% | 85.58% | 86.11% | 90.23% | -| SVTR-Tiny-8P | 95.93% | 95.62% | 95.33% | 93.89% | 84.32% | 80.55% | 94.33% | 90.57% | 86.20% | 86.46% | 90.32% | +| **model name** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **average** | +|:--------------:| :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: |:-----------:| +| SVTR-Tiny | 95.70% | 95.50% | 95.33% | 93.99% | 83.60% | 79.83% | 94.70% | 91.96% | 85.58% | 86.11% | 90.23% | +| SVTR-Tiny-8P | 95.93% | 95.62% | 95.33% | 93.89% | 84.32% | 80.55% | 94.33% | 90.57% | 86.20% | 86.46% | 90.32% |
@@ -377,9 +377,9 @@ After training, evaluation results on the benchmark test set are as follows, whe
-| **Model** | **Language** | **Context** | **Scene** | **Web** | **Document** | **Train T.** | **FPS** | **Recipe** | **Download** | -| :-----: | :-----: | :--------: | :--------: | :--------: | :--------: | :---------: | :--------: | :---------: | :-----------: | -| SVTR-Tiny | Chinese | D910x4-MS1.10-G | 65.93% | 69.64% | 98.01% | 647 s/epoch | 1580 | [svtr_tiny_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4-3e495768.mindir) | +| **model name** | **cards** | **batch size** | **language** | **jit level** | **graph compile** | **ms/step** | **img/s** | **scene** | **web** | **document** | **recipe** | **download** | +|:--------------:|:---------:|:--------------:| :--------: |:-------------:|:-----------------:|:---------:|:-------:|:------------:|:-----------:|:---------:|:----------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| SVTR-Tiny | 4 | 256 | Chinese | O2 | 235.1 s| 37.75 | 1580 | 65.93% | 69.64% | 98.01% | [svtr_tiny_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4-3e495768.mindir) |
### Training with Custom Datasets diff --git a/configs/rec/svtr/README_CN.md b/configs/rec/svtr/README_CN.md index 943134492..839b28cf7 100644 --- a/configs/rec/svtr/README_CN.md +++ b/configs/rec/svtr/README_CN.md @@ -40,24 +40,23 @@ Table Format: 根据我们的实验，在公开基准数据集（IC03，IC13，IC15，IIIT，SVT，SVTP，CUTE）上的评估结果如下： -#### 在采用图模式的ascend 910上测试性能 +#### 在采用图模式的ascend 910上实验结果，mindspore版本为2.3.1
- | **Model** | **Cards** | **Avg Accuracy** | **Batch size** | **graph compile** |**jit level** | **Step Time** | **FPS** | **Recipe** | **Download** | - | :-----: |:---------:| :--------------: |:--------------:|:-----------------:|:-----------------:|:-------------:| :--------: | :--------: |:----------: | - | SVTR-Tiny | 4 | 90.23% | 512 | 226.86 s |O2| 49.38 ms/step | 4560 | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3-86ece8c8.mindir) | - | SVTR-Tiny-8P | 8 | 90.32% | 512 | 226.86 s |O2| 55.16 ms/step | 9840 | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6.ckpt) \| [mindir](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6-255191ef.mindir) | +| **model name** | **cards** | **batch size** |**jit level** | **graph compile** | **ms/step** | **img/s** | **avg accuracy** | **recipe** | **download** | +|:--------------:|:---------:|:--------------:|:----------------:|:-----------------:|:-----------------:|:-------------:|:---------:|:---------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| SVTR-Tiny | 4 | 512 |O2| 226.86 s | 49.38 ms/step | 4560 | 90.23% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny-950be1c3-86ece8c8.mindir) | +| SVTR-Tiny-8P | 8 | 512 |O2| 226.86 s | 55.16 ms/step | 9840 | 90.32% | [yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_8p.yaml) | [ckpt](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6.ckpt) \| [mindir](https://download-mindspore.osinfra.cn/toolkits/mindocr/svtr/svtr_tiny_8p-0afc75d6-255191ef.mindir) |
在各个基准数据集上的准确率
-| **模型** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **平均准确率** | -|:------------:|:------------:|:------------:|:------------:|:-------------:|:-------------:|:-------------:|:---------------:|:-------:|:--------:|:----------:|:---------:| -| SVTR-Tiny | 95.70% | 95.50% | 95.33% | 93.99% | 83.60% | 79.83% | 94.70% | 91.96% | 85.58% | 86.11% | 90.23% | -| SVTR-Tiny-8P | 95.93% | 95.62% | 95.33% | 93.89% | 84.32% | 80.55% | 94.33% | 90.57% | 86.20% | 86.46% | 90.32% | - +| **model name** | **IC03_860** | **IC03_867** | **IC13_857** | **IC13_1015** | **IC15_1811** | **IC15_2077** | **IIIT5k_3000** | **SVT** | **SVTP** | **CUTE80** | **average** | +|:--------------:| :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: | :------: |:-----------:| +| SVTR-Tiny | 95.70% | 95.50% | 95.33% | 93.99% | 83.60% | 79.83% | 94.70% | 91.96% | 85.58% | 86.11% | 90.23% | +| SVTR-Tiny-8P | 95.93% | 95.62% | 95.33% | 93.89% | 84.32% | 80.55% | 94.33% | 90.57% | 86.20% | 86.46% | 90.32% |
@@ -374,9 +373,9 @@ mpirun --allow-run-as-root -n 4 python tools/train.py --config configs/rec/svtr/
-| **模型** | **语种** | **环境配置** | **街景类** | **网页类** | **文档类** | **训练时间** | **FPS** | **配置文件** | **模型权重下载** | -| :-----: | :-----: | :-------: | :--------: | :--------: | :--------: | :---------: |:--------: | :---------: | :-----------: | -| SVTR-Tiny | 中文 | D910x4-MS1.10-G | 65.93% | 69.64% | 98.01% | 647 s/epoch | 1580 | [svtr_tiny_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4.ckpt) \| [mindir]() | +| **model name** | **cards** | **batch size** | **language** | **jit level** | **graph compile** | **ms/step** | **img/s** | **scene** | **web** | **document** | **recipe** | **download** | +|:--------------:|:---------:|:--------------:| :--------: |:-------------:|:-----------------:|:---------:|:-------:|:------------:|:-----------:|:---------:|:----------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| +| SVTR-Tiny | 4 | 256 | Chinese | O2 | 235.1 s| 65.93% | 37.75 | 1580 | 69.64% | 98.01% | [svtr_tiny_ch.yaml](https://github.com/mindspore-lab/mindocr/blob/main/configs/rec/svtr/svtr_tiny_ch.yaml) | [ckpt](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4.ckpt) \| [mindir](https://download.mindspore.cn/toolkits/mindocr/svtr/svtr_tiny_ch-2ee6ade4-3e495768.mindir) |
### 使用自定义数据集进行训练