add

dvlab-research · Apr 14, 2021 · bb5ba7e · bb5ba7e
1 parent ac9c354
commit bb5ba7e
Show file tree

Hide file tree

Showing 43 changed files with 731,295 additions and 22 deletions.
diff --git a/README.md b/README.md
@@ -34,10 +34,10 @@ pip install -r requirements.txt
 ```
 
 **Dataset Preparation**
-* [CIFAR-10-LT, CIFAR-100-LT](https://www.cs.toronto.edu/~kriz/cifar.html)
-* [ImageNet_LT](http://image-net.org/index)
+* [CIFAR-10 & CIFAR-100](https://www.cs.toronto.edu/~kriz/cifar.html)
+* [ImageNet](http://image-net.org/index)
 * [iNaturalist 2018](https://github.com/visipedia/inat_comp/tree/master/2018)
-* [Places_LT](http://places2.csail.mit.edu/download.html)
+* [Places](http://places2.csail.mit.edu/download.html)
 
 Change the `data_path` in `config/*/*.yaml` accordingly.
 
@@ -47,7 +47,7 @@ Change the `data_path` in `config/*/*.yaml` accordingly.
 
 To train a model for Stage-1 with *mixup*, run:
 
-(one GPU for CIFAR-10-LT, CIFAR-100-LT, four GPUs for ImageNet-LT, iNaturalist 2018, Places-LT)
+(one GPU for CIFAR-10-LT & CIFAR-100-LT, four GPUs for ImageNet-LT, iNaturalist 2018, and Places-LT)
 
 ```
 python train_stage1.py --cfg ./config/DATASETNAME/DATASETNAME_ARCH_stage1_mixup.yaml
@@ -64,6 +64,19 @@ To train a model for Stage-2 with *one GPU*, run:
 ```
 python train_stage2.py --cfg ./config/DATASETNAME/DATASETNAME_ARCH_stage2_mislas.yaml resume /path/to/checkpoint/stage1
 ```
+
+The saved folder (including logs and checkpoints) is organized as follows.
+```
+MiSLAS
+├── saved
+│   ├── model name + date
+│   │   ├── ckps
+│   │   │   ├── current.pth.tar
+│   │   │   └── model_best.pth.tar
+│   │   └── logs
+│   │   │   └── model name.txt
+│   ...   
+```
 ## Evaluation
 
 To evaluate a pre-trained model, run:
@@ -81,23 +94,23 @@ python eval.py --cfg ./config/DATASETNAME/DATASETNAME_ARCH_stage2_mislas.yaml re
 
 | Dataset              | Top-1 Accuracy | ECE (15 bins) | Model |
 | -------------------- | -------------- | ------------- | ----- |
-| CIFAR-10-LT   IF=10  | 87.6%          | 11.9%         | link  |
-| CIFAR-10-LT   IF=50  | 78.1%          | 2.49%         | link  |
-| CIFAR-10-LT   IF=100 | 72.8%          | 2.14%         | link  |
-| CIFAR-100-LT IF=10   | 59.1%          | 5.24%         | link  |
-| CIFAR-100-LT IF=50   | 45.4%          | 4.33%         | link  |
-| CIFAR-100-LT IF=100  | 39.5%          | 8.82%         | link  |
+| CIFAR-10-LT   IF=10  | 87.6%          | 11.9%         | [link](https://drive.google.com/file/d/1dV1hchsIR5kTSqSOhdEs6nnXApcH5wEG/view?usp=sharing)  |
+| CIFAR-10-LT   IF=50  | 78.1%          | 2.49%         | [link](https://drive.google.com/file/d/1LoczjQRK20u_HpFMLmzeT0pVCp3V-gyf/view?usp=sharing)  |
+| CIFAR-10-LT   IF=100 | 72.8%          | 2.14%         | [link](https://drive.google.com/file/d/1TFetlV4MT4zjKEAPKcZuzmY2Dgtcqmsd/view?usp=sharing)  |
+| CIFAR-100-LT IF=10   | 59.1%          | 5.24%         | [link](https://drive.google.com/file/d/1BmLjPReBoH6LJwl5x8_zSPnm1f6N_Cp0/view?usp=sharing)  |
+| CIFAR-100-LT IF=50   | 45.4%          | 4.33%         | [link](https://drive.google.com/file/d/1l0LfZozJxWgzKp2IgM9mSpfwjTsIC-Mg/view?usp=sharing)  |
+| CIFAR-100-LT IF=100  | 39.5%          | 8.82%         | [link](https://drive.google.com/file/d/15dHVdkI8J-oKkeQqyj6FtrHtIpO_TYfq/view?usp=sharing)  |
 
 * Stage-2 (*MiSLAS*):
 
 | Dataset              | Top-1 Accuracy | ECE (15 bins) | Model |
 | -------------------- | -------------- | ------------- | ----- |
-| CIFAR-10-LT   IF=10  | 90.0%          | 1.20%         | link  |
-| CIFAR-10-LT   IF=50  | 85.7%          | 2.01%         | link  |
-| CIFAR-10-LT   IF=100 | 82.5%          | 3.66%         | link  |
-| CIFAR-100-LT IF=10   | 63.2%          | 1.73%         | link  |
-| CIFAR-100-LT IF=50   | 52.3%          | 2.47%         | link  |
-| CIFAR-100-LT IF=100  | 47.0%          | 4.83%         | link  |
+| CIFAR-10-LT   IF=10  | 90.0%          | 1.20%         | [link](https://drive.google.com/file/d/1iST8Tr2LQ8nIjTNT1CKiQ-1T-RKxAvqr/view?usp=sharing)  |
+| CIFAR-10-LT   IF=50  | 85.7%          | 2.01%         | [link](https://drive.google.com/file/d/15bfA7uJsyM8eTwoptwp452kStk6FYT7v/view?usp=sharing)  |
+| CIFAR-10-LT   IF=100 | 82.5%          | 3.66%         | [link](https://drive.google.com/file/d/1KOTkjTOhIP5UOhqvHGJzEqq4_kQGKSJY/view?usp=sharing)  |
+| CIFAR-100-LT IF=10   | 63.2%          | 1.73%         | [link](https://drive.google.com/file/d/1N2ai-l1hsbXTp_25Hoh5BSoAmR1_0UVD/view?usp=sharing)  |
+| CIFAR-100-LT IF=50   | 52.3%          | 2.47%         | [link](https://drive.google.com/file/d/1Z2nukCMTG0cMmGXzZip3zIwv2WB5cOiZ/view?usp=sharing)  |
+| CIFAR-100-LT IF=100  | 47.0%          | 4.83%         | [link](https://drive.google.com/file/d/1bX3eM-hlxGvEGuHBcfNhuz6VNp32Y0IQ/view?usp=sharing)  |
 
 *Note: To obtain better performance, we highly recommend changing the weight decay 2e-4 to 5e-4 on CIFAR-LT.*
 
@@ -107,17 +120,17 @@ python eval.py --cfg ./config/DATASETNAME/DATASETNAME_ARCH_stage2_mislas.yaml re
 
 | Dataset     | Arch       | Top-1 Accuracy | ECE (15 bins) | Model |
 | ----------- | ---------- | -------------- | ------------- | ----- |
-| ImageNet-LT | ResNet-50  | 45.5%          | 7.98%         | link  |
-| iNa'2018    | ResNet-50  | 66.9%          | 5.37%         | link  |
-| Places-LT   | ResNet-152 | 29.4%          | 16.7%         | link  |
+| ImageNet-LT | ResNet-50  | 45.5%          | 7.98%         | [link](https://drive.google.com/file/d/1QKVnK7n75q465ppf7wkK4jzZvZJE_BPi/view?usp=sharing)  |
+| iNa'2018    | ResNet-50  | 66.9%          | 5.37%         | [link](https://drive.google.com/file/d/1wvj-cITz8Ps1TksLHi_KoGsq9CecXcVt/view?usp=sharing)  |
+| Places-LT   | ResNet-152 | 29.4%          | 16.7%         | [link](https://drive.google.com/file/d/1Tx-tY5Y8_-XuGn9ZdSxtAm0onOsKWhUH/view?usp=sharing)  |
 
 * Stage-2 (*MiSLAS*):
 
 | Dataset     | Arch       | Top-1 Accuracy | ECE (15 bins) | Model |
 | ----------- | ---------- | -------------- | ------------- | ----- |
-| ImageNet-LT | ResNet-50  | 52.7%          | 1.78%         | link  |
-| iNa'2018    | ResNet-50  | 71.6%          | 7.67%         | link  |
-| Places-LT   | ResNet-152 | 40.4%          | 3.41%         | link  |
+| ImageNet-LT | ResNet-50  | 52.7%          | 1.78%         | [link](https://drive.google.com/file/d/1ofJKlUJZQjjkoFU9MLI08UP2uBvywRgF/view?usp=sharing)  |
+| iNa'2018    | ResNet-50  | 71.6%          | 7.67%         | [link](https://drive.google.com/file/d/1crOo3INxqkz8ZzKZt9pH4aYb3-ep4lo-/view?usp=sharing)  |
+| Places-LT   | ResNet-152 | 40.4%          | 3.41%         | [link](https://drive.google.com/file/d/1DgL0aN3UadI3UoHU6TO7M6UD69QgvnbT/view?usp=sharing)  |
 
 ## <a name="Citation"></a>Citation
 

diff --git a/config/cifar10/cifar10_imb001_stage1_mixup.yaml b/config/cifar10/cifar10_imb001_stage1_mixup.yaml
@@ -0,0 +1,50 @@
+name: cifar10_imb001_stage1_mixup
+print_freq: 40
+workers: 16
+log_dir: 'logs'
+model_dir: 'ckps'
+
+# dataset & model setting
+dataset: 'cifar10'
+data_path: './data/cifar10'
+num_classes: 10
+imb_factor: 0.01
+backbone: 'resnet32_fe'
+resume: ''
+head_class_idx:
+  - 0
+  - 3
+med_class_idx:
+  - 3
+  - 7
+tail_class_idx:
+  - 7
+  - 10
+
+
+# distributed training
+deterministic: False
+distributed: False
+gpu: null
+world_size: -1
+rank: -1
+dist_url: 'tcp://224.66.41.62:23456'
+dist_backend: 'nccl'
+multiprocessing_distributed: False
+
+
+
+# Train
+mode: 'stage1'
+lr: 0.1
+batch_size: 128
+weight_decay: 2e-4
+num_epochs: 200
+momentum: 0.9
+cos: False
+mixup: True
+alpha: 1.0
+
+
+
+
diff --git a/config/cifar10/cifar10_imb001_stage2_mislas.yaml b/config/cifar10/cifar10_imb001_stage2_mislas.yaml
@@ -0,0 +1,54 @@
+name: cifar10_imb001_stage2_mislas
+print_freq: 40
+workers: 16
+log_dir: 'logs'
+model_dir: 'ckps'
+
+# dataset & model setting
+dataset: 'cifar10'
+data_path: './data/cifar10'
+num_classes: 10
+imb_factor: 0.01
+backbone: 'resnet32_fe'
+resume: 'Path/to/Stage1_checkpoint.pth.tar'
+head_class_idx:
+  - 0
+  - 3
+med_class_idx:
+  - 3
+  - 7
+tail_class_idx:
+  - 7
+  - 10
+
+
+# distributed training
+deterministic: False
+distributed: False
+gpu: null
+world_size: -1
+rank: -1
+dist_url: 'tcp://224.66.41.62:23456'
+dist_backend: 'nccl'
+multiprocessing_distributed: False
+
+
+
+# Train
+mode: 'stage2'
+smooth_head: 0.3
+smooth_tail: 0.0
+shift_bn: False
+lr_factor: 0.5
+lr: 0.1
+batch_size: 128
+weight_decay: 2e-4
+num_epochs: 10
+momentum: 0.9
+mixup: False
+alpha: null
+
+
+
+
+
diff --git a/config/cifar10/cifar10_imb002_stage1_mixup.yaml b/config/cifar10/cifar10_imb002_stage1_mixup.yaml
@@ -0,0 +1,50 @@
+name: cifar10_imb002_stage1_mixup
+print_freq: 40
+workers: 16
+log_dir: 'logs'
+model_dir: 'ckps'
+
+# dataset & model setting
+dataset: 'cifar10'
+data_path: './data/cifar10'
+num_classes: 10
+imb_factor: 0.02
+backbone: 'resnet32_fe'
+resume: ''
+head_class_idx:
+  - 0
+  - 3
+med_class_idx:
+  - 3
+  - 7
+tail_class_idx:
+  - 7
+  - 10
+
+
+# distributed training
+deterministic: False
+distributed: False
+gpu: null
+world_size: -1
+rank: -1
+dist_url: 'tcp://224.66.41.62:23456'
+dist_backend: 'nccl'
+multiprocessing_distributed: False
+
+
+
+# Train
+mode: 'stage1'
+lr: 0.1
+batch_size: 128
+weight_decay: 2e-4
+num_epochs: 200
+momentum: 0.9
+cos: False
+mixup: True
+alpha: 1.0
+
+
+
+
diff --git a/config/cifar10/cifar10_imb002_stage2_mislas.yaml b/config/cifar10/cifar10_imb002_stage2_mislas.yaml
@@ -0,0 +1,54 @@
+name: cifar10_imb002_stage2_mislas
+print_freq: 40
+workers: 16
+log_dir: 'logs'
+model_dir: 'ckps'
+
+# dataset & model setting
+dataset: 'cifar10'
+data_path: './data/cifar10'
+num_classes: 10
+imb_factor: 0.02
+backbone: 'resnet32_fe'
+resume: 'Path/to/Stage1_checkpoint.pth.tar'
+head_class_idx:
+  - 0
+  - 3
+med_class_idx:
+  - 3
+  - 7
+tail_class_idx:
+  - 7
+  - 10
+
+
+# distributed training
+deterministic: False
+distributed: False
+gpu: null
+world_size: -1
+rank: -1
+dist_url: 'tcp://224.66.41.62:23456'
+dist_backend: 'nccl'
+multiprocessing_distributed: False
+
+
+
+# Train
+mode: 'stage2'
+smooth_head: 0.2
+smooth_tail: 0.0
+shift_bn: False
+lr_factor: 0.2
+lr: 0.1
+batch_size: 128
+weight_decay: 2e-4
+num_epochs: 10
+momentum: 0.9
+mixup: False
+alpha: null
+
+
+
+
+
diff --git a/config/cifar10/cifar10_imb01_stage1_mixup.yaml b/config/cifar10/cifar10_imb01_stage1_mixup.yaml
@@ -0,0 +1,50 @@
+name: cifar10_imb01_stage1_mixup
+print_freq: 40
+workers: 16
+log_dir: 'logs'
+model_dir: 'ckps'
+
+# dataset & model setting
+dataset: 'cifar10'
+data_path: './data/cifar10'
+num_classes: 10
+imb_factor: 0.1
+backbone: 'resnet32_fe'
+resume: ''
+head_class_idx:
+  - 0
+  - 3
+med_class_idx:
+  - 3
+  - 7
+tail_class_idx:
+  - 7
+  - 10
+
+
+# distributed training
+deterministic: False
+distributed: False
+gpu: null
+world_size: -1
+rank: -1
+dist_url: 'tcp://224.66.41.62:23456'
+dist_backend: 'nccl'
+multiprocessing_distributed: False
+
+
+
+# Train
+mode: 'stage1'
+lr: 0.1
+batch_size: 128
+weight_decay: 2e-4
+num_epochs: 200
+momentum: 0.9
+cos: False
+mixup: True
+alpha: 1.0
+
+
+
+