[Feature] EfficientNets NoisyStudent & L2. (#1122)

* add mvit 21k

* add mvit 21k

* add effnet noisy student

* Revert "add mvit 21k"

This reverts commit f51067c.

* revert mvit pr

* update link and readme

* update readme

* update l2 link

* update link

Co-authored-by: Ezra-Yu <[email protected]>

.gitignore

@@ -136,3 +136,10 @@ venv.bak/
 
 # slurm
 *.out
+
+# tensorflow
+*.tar.gz
+checkpoint
+model_params.txt
+*.ckpt*
+results.txt

configs/_base_/models/efficientnet_l2.py

@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='l2'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=5504,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))

configs/efficientnet/efficientnet-l2_8xb32_in1k-475px.py

@@ -0,0 +1,24 @@
+_base_ = [
+    '../_base_/models/efficientnet_l2.py',
+    '../_base_/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py',
+    '../_base_/default_runtime.py',
+]
+
+# dataset settings
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='EfficientNetRandomCrop', scale=475),
+    dict(type='RandomFlip', prob=0.5, direction='horizontal'),
+    dict(type='PackClsInputs'),
+]
+
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='EfficientNetCenterCrop', crop_size=475),
+    dict(type='PackClsInputs'),
+]
+
+train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
+val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
+test_dataloader = dict(dataset=dict(pipeline=test_pipeline))

configs/efficientnet/efficientnet-l2_8xb8_in1k.py

@@ -0,0 +1,24 @@
+_base_ = [
+    '../_base_/models/efficientnet_l2.py',
+    '../_base_/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py',
+    '../_base_/default_runtime.py',
+]
+
+# dataset settings
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='EfficientNetRandomCrop', scale=800),
+    dict(type='RandomFlip', prob=0.5, direction='horizontal'),
+    dict(type='PackClsInputs'),
+]
+
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='EfficientNetCenterCrop', crop_size=800),
+    dict(type='PackClsInputs'),
+]
+
+train_dataloader = dict(batch_size=8, dataset=dict(pipeline=train_pipeline))
+val_dataloader = dict(batch_size=8, dataset=dict(pipeline=test_pipeline))
+test_dataloader = dict(batch_size=8, dataset=dict(pipeline=test_pipeline))