CarFusion generated by your guidance is not consistent with your annotations.

[kltrock]

I followed your guidance to generate CarFusion.

First, I downloaded all the zip files from CarFusion website and extract them into:
-train: car_butler1 car_butler2 car_craig1 car_craig2 car_fifth1 car_fifth2 car_morewood1 car_morewood2
-test: car_penn1 car_penn2

Second, I used the official tools to convert the annotations to COCO format, resulting in car_keypoints_test.json, car_keypoints_train.json.

Finally, I renamed all the jpg files using your rename_carfusion_image.py.

However, the images obtained are not consistent with your annotation files. For example, in your annotation files, there is an image with name 'car/1500801201.jpg' which is missing in the generated images.

[kltrock]

Could you please provide more details on how to obtain the CarFusion correctly?

[luminxu]

Thanks for your question. I am wondering whether all the annotations are inconsistent with the generated images or only some of them are missing. Would you mind providing some examples of the names of the generated images?

[kltrock]

I think all the annotations are inconsistent. As I have described earlier, I downloaded the Carfusion datasets, which resulted in 10 folders (car_butler1, car_butler2, car_craig1, car_craig2, car_fifth1, car_fifth2, car_morewood1, car_morewood2, car_penn1, car_penn2).

However, following the official tools, the image_id of each image is defined by 'loop', 'video_id', 'frame_id'. The 'loop' variable is loosely defined by the order of the 10 folders. The highest value of the 'loop' variable is 10, so the image_id would be 10xxxxxxxx.jpg, 9xxxxxxxx.jpg, etc. However, in your annotations, there are a lot of annotations with image names 15xxxxxxxx.jpg, 14xxxxxxxx.jpg, etc.

Could you please provide more details or a script to correctly generate the dataset?

[A465539338]

I found the same problem as kltrock. And there is another question for me. It seems all the train and test images from CarFusion dataset shares the only category car which means no annotation for bus and suv. Could you please tell more about how to arrage this CarFusion dataset?

[Annbless]

Thanks for your question. I am wondering whether all the annotations are inconsistent with the generated images or only some of them are missing. Would you mind providing some examples of the names of the generated images?

It seems that all the annotations are inconsistent with the generated images. I have noticed that the names of the images generated following the guidance have 9-digit numbers while the names provided in the annotations have 10-digit numbers.

[worklikerlei]

I found the same problem as kltrock. And there is another question for me. It seems all the train and test images from CarFusion dataset shares the only category car which means no annotation for bus and suv. Could you please tell more about how to arrage this CarFusion dataset?

Hi, I have also encountered this problem. Have you solved it?

[worklikerlei]

Hello, have you solved the problem of this dataset？

[kltrock]

Yes. I check every instance in their annotation files and find the corresponding image accordingly.

[worklikerlei]

Yes. I check every instance in their annotation files and find the corresponding image accordingly.

Thank you very much for your reply. following the https://github.com/dineshreddy91/carfusion_to_coco/blob/master/carfusion2coco.py#L164
It seems all the train and test images from CarFusion dataset shares the only category car which means no annotation for bus and suv. Could you give me some help. Thank you very much!

[worklikerlei]

Hello, for the mp100 dataset in the Pose-for-Everything task, the carfusion datasets image does not match. Can you provide some ideas to help me? Thank you very much.

…

-- | | fengzaifei12 | | ***@***.*** | 在 2023-02-02 16:48:26，"kltrock" ***@***.***> 写道： Yes. I check every instance in their annotation files and find the corresponding image accordingly. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[kltrock]

I suggest you examine every instance in the annotation files of Carfusion and Mp100. You can match one instance in Carfusion with one instance in Mp100 if they have identical information (keypoint positions, number of key points, etc). After matching, you can trace back the images in the Carfusion.

[worklikerlei]

There are more than 30000 pictures generated by carfusion_ to_ coco. It is very difficult for each annotaiton to find pictures. What are the skills about this?

…

-- | | fengzaifei12 | | ***@***.*** | 在 2023-02-09 15:54:48，"kltrock" ***@***.***> 写道： I suggest you examine every instance in the annotation files of Carfusion and Mp100. You can match one instance in Carfusion with one instance in Mp100 if they have identical information (keypoint positions, number of key points, etc). After matching, you can trace back the images in the Carfusion. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[kltrock]

You can write a script for it. It would be fast.

[worklikerlei]

If it is convenient, can we add a WeChat chat for a few minutes without much interruption, thank you very much！my wechat is 15811013659.

…

-- | | fengzaifei12 | | ***@***.*** | 在 2023-02-09 15:54:48，"kltrock" ***@***.***> 写道： I suggest you examine every instance in the annotation files of Carfusion and Mp100. You can match one instance in Carfusion with one instance in Mp100 if they have identical information (keypoint positions, number of key points, etc). After matching, you can trace back the images in the Carfusion. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>