[Fix] Remove fp16 folder in configs. (open-mmlab#1031)

* remove fp16 folder * remove fp16 in docs * fix some typos * fix some typos * fix fp16 in yml
bowenroom · Nov 15, 2021 · 008856a · 008856a
1 parent d4ec8ef
commit 008856a
Show file tree

Hide file tree

Showing 25 changed files with 119 additions and 151 deletions.
diff --git a/.dev/batch_test_list.py b/.dev/batch_test_list.py
@@ -116,8 +116,8 @@
 ]
 fp16 = [
     dict(
-        config='configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py', # noqa
-        checkpoint='deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes-cc58bc8d.pth', # noqa
+        config='configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py', # noqa
+        checkpoint='deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes_20200717_230920-f1104f4b.pth', # noqa
         eval='mIoU',
         metric=dict(mIoU=80.46),
     )

diff --git a/.dev/batch_train_list.txt b/.dev/batch_train_list.txt
@@ -15,5 +15,5 @@ configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_80k_cityscapes.py
 configs/deeplabv3plus/deeplabv3plus_r50-d8_769x769_80k_cityscapes.py
 configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py
 configs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k.py
-configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py
+configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py
 configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py
diff --git a/.dev/benchmark_evaluation.sh b/.dev/benchmark_evaluation.sh
@@ -35,7 +35,7 @@ echo 'configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py' &
 GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 tools/slurm_test.sh $PARTITION upernet_vit-b16_ln_mln_512x512_160k_ade20k configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py $CHECKPOINT_DIR/upernet_vit-b16_ln_mln_512x512_160k_ade20k-f444c077.pth --eval mIoU --work-dir work_dirs/benchmark_evaluation/upernet_vit-b16_ln_mln_512x512_160k_ade20k --options dist_params.port=28186 &
 echo 'configs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k.py' &
 GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 tools/slurm_test.sh $PARTITION upernet_deit-s16_ln_mln_512x512_160k_ade20k configs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k.py $CHECKPOINT_DIR/upernet_deit-s16_ln_mln_512x512_160k_ade20k-c0cd652f.pth --eval mIoU --work-dir work_dirs/benchmark_evaluation/upernet_deit-s16_ln_mln_512x512_160k_ade20k --options dist_params.port=28187 &
-echo 'configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py' &
-GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 tools/slurm_test.sh $PARTITION deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py $CHECKPOINT_DIR/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes-cc58bc8d.pth --eval mIoU --work-dir work_dirs/benchmark_evaluation/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes --options dist_params.port=28188 &
+echo 'configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py' &
+GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 tools/slurm_test.sh $PARTITION deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py $CHECKPOINT_DIR/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes-cc58bc8d.pth --eval mIoU --work-dir work_dirs/benchmark_evaluation/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes --options dist_params.port=28188 &
 echo 'configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py' &
 GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 tools/slurm_test.sh $PARTITION upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py $CHECKPOINT_DIR/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K_20210531_112542-e380ad3e.pth --eval mIoU --work-dir work_dirs/benchmark_evaluation/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K --options dist_params.port=28189 &
diff --git a/.dev/benchmark_train.sh b/.dev/benchmark_train.sh
@@ -34,7 +34,7 @@ echo 'configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py' &
 GPUS=8  GPUS_PER_NODE=8  CPUS_PER_TASK=2 ./tools/slurm_train.sh $PARTITION upernet_vit-b16_ln_mln_512x512_160k_ade20k configs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k.py --options checkpoint_config.max_keep_ckpts=1 dist_params.port=24742 --work-dir work_dirs/vit/upernet_vit-b16_ln_mln_512x512_160k_ade20k >/dev/null &
 echo 'configs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k.py' &
 GPUS=8  GPUS_PER_NODE=8  CPUS_PER_TASK=2 ./tools/slurm_train.sh $PARTITION upernet_deit-s16_ln_mln_512x512_160k_ade20k configs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k.py --options checkpoint_config.max_keep_ckpts=1 dist_params.port=24743 --work-dir work_dirs/vit/upernet_deit-s16_ln_mln_512x512_160k_ade20k >/dev/null &
-echo 'configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py' &
-GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 ./tools/slurm_train.sh $PARTITION deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes configs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes.py --options checkpoint_config.max_keep_ckpts=1 dist_params.port=24744 --work-dir work_dirs/fp16/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes >/dev/null &
+echo 'configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py' &
+GPUS=4  GPUS_PER_NODE=4  CPUS_PER_TASK=2 ./tools/slurm_train.sh $PARTITION deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes configs/deeplabv3plus/deeplabv3plus_r101-d8_fp16_512x1024_80k_cityscapes.py --options checkpoint_config.max_keep_ckpts=1 dist_params.port=24744 --work-dir work_dirs/deeplabv3plus/deeplabv3plus_r101-d8_512x1024_80k_fp16_cityscapes >/dev/null &
 echo 'configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py' &
 GPUS=8  GPUS_PER_NODE=8  CPUS_PER_TASK=2 ./tools/slurm_train.sh $PARTITION upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K configs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py --options checkpoint_config.max_keep_ckpts=1 dist_params.port=24745 --work-dir work_dirs/swin/upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K >/dev/null &
diff --git a/.dev/md2yml.py b/.dev/md2yml.py
@@ -201,7 +201,7 @@ def parse_md(md_file):
                             'batch size':
                             1,
                             'mode':
-                            'FP32',
+                            'FP32' if 'fp16' not in config else 'FP16',
                             'resolution':
                             f'({crop_size[0]},{crop_size[1]})'
                         }]

diff --git a/README.md b/README.md
@@ -73,7 +73,6 @@ Supported methods:
 - [x] [UNet (MICCAI'2016/Nat. Methods'2019)](configs/unet)
 - [x] [PSPNet (CVPR'2017)](configs/pspnet)
 - [x] [DeepLabV3 (ArXiv'2017)](configs/deeplabv3)
-- [x] [Mixed Precision (FP16) Training (ArXiv'2017)](configs/fp16)
 - [x] [BiSeNetV1 (ECCV'2018)](configs/bisenetv1)
 - [x] [PSANet (ECCV'2018)](configs/psanet)
 - [x] [DeepLabV3+ (CVPR'2018)](configs/deeplabv3plus)

diff --git a/README_zh-CN.md b/README_zh-CN.md
@@ -72,7 +72,6 @@ MMSegmentation 是一个基于 PyTorch 的语义分割开源工具箱。它是 O
 - [x] [UNet (MICCAI'2016/Nat. Methods'2019)](configs/unet)
 - [x] [PSPNet (CVPR'2017)](configs/pspnet)
 - [x] [DeepLabV3 (ArXiv'2017)](configs/deeplabv3)
-- [x] [Mixed Precision (FP16) Training (ArXiv'2017)](configs/fp16)
 - [x] [BiSeNetV1 (ECCV'2018)](configs/bisenetv1)
 - [x] [PSANet (ECCV'2018)](configs/psanet)
 - [x] [DeepLabV3+ (CVPR'2018)](configs/deeplabv3plus)

diff --git a/configs/bisenetv2/bisenetv2.yml b/configs/bisenetv2/bisenetv2.yml
@@ -75,7 +75,7 @@ Models:
       hardware: V100
       backend: PyTorch
       batch size: 1
-      mode: FP32
+      mode: FP16
       resolution: (1024,1024)
     memory (GB): 5.77
   Results:

diff --git a/configs/deeplabv3/README.md b/configs/deeplabv3/README.md
@@ -39,6 +39,7 @@
 | DeepLabV3 | R-18-D8         | 512x1024  |   80000 | 1.7      | 13.78          | 76.70 |         78.27 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r18-d8_512x1024_80k_cityscapes.py)         | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r18-d8_512x1024_80k_cityscapes/deeplabv3_r18-d8_512x1024_80k_cityscapes_20201225_021506-23dffbe2.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r18-d8_512x1024_80k_cityscapes/deeplabv3_r18-d8_512x1024_80k_cityscapes-20201225_021506.log.json)                                 |
 | DeepLabV3 | R-50-D8         | 512x1024  |   80000 | -        | -              | 79.32 |         80.57 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r50-d8_512x1024_80k_cityscapes.py)         | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x1024_80k_cityscapes/deeplabv3_r50-d8_512x1024_80k_cityscapes_20200606_113404-b92cfdd4.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x1024_80k_cityscapes/deeplabv3_r50-d8_512x1024_80k_cityscapes_20200606_113404.log.json)                                 |
 | DeepLabV3 | R-101-D8        | 512x1024  |   80000 | -        | -              | 80.20 |         81.21 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r101-d8_512x1024_80k_cityscapes.py)        | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x1024_80k_cityscapes/deeplabv3_r101-d8_512x1024_80k_cityscapes_20200606_113503-9e428899.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x1024_80k_cityscapes/deeplabv3_r101-d8_512x1024_80k_cityscapes_20200606_113503.log.json)                             |
+| DeepLabV3 (FP16)  | R-101-D8 | 512x1024  |   80000 | 5.75     | 3.86           | 80.48 |             - | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes.py)     | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes_20200717_230920-774d9cec.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes_20200717_230920.log.json)                 |
 | DeepLabV3 | R-18-D8         | 769x769   |   80000 | 1.9      | 5.55           | 76.60 |         78.26 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r18-d8_769x769_80k_cityscapes.py)          | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r18-d8_769x769_80k_cityscapes/deeplabv3_r18-d8_769x769_80k_cityscapes_20201225_021506-6452126a.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r18-d8_769x769_80k_cityscapes/deeplabv3_r18-d8_769x769_80k_cityscapes-20201225_021506.log.json)                                     |
 | DeepLabV3 | R-50-D8         | 769x769   |   80000 | -        | -              | 79.89 |         81.06 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r50-d8_769x769_80k_cityscapes.py)          | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_769x769_80k_cityscapes/deeplabv3_r50-d8_769x769_80k_cityscapes_20200606_221338-788d6228.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_769x769_80k_cityscapes/deeplabv3_r50-d8_769x769_80k_cityscapes_20200606_221338.log.json)                                     |
 | DeepLabV3 | R-101-D8        | 769x769   |   80000 | -        | -              | 79.67 |         80.81 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3/deeplabv3_r101-d8_769x769_80k_cityscapes.py)         | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_769x769_80k_cityscapes/deeplabv3_r101-d8_769x769_80k_cityscapes_20200607_013353-60e95418.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_769x769_80k_cityscapes/deeplabv3_r101-d8_769x769_80k_cityscapes_20200607_013353.log.json)                                 |
@@ -102,3 +103,7 @@
 | DeepLabV3 | R-101-D8 | 512x512   |  160000 | -        | -              | 41.82 |         42.49 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3_r101-d8_512x512_4x4_160k_coco-stuff164k.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_4x4_160k_coco-stuff164k/deeplabv3_r101-d8_512x512_4x4_160k_coco-stuff164k_20210709_155402-f035acfd.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_4x4_160k_coco-stuff164k/deeplabv3_r101-d8_512x512_4x4_160k_coco-stuff164k_20210709_155402.log.json) |
 | DeepLabV3 | R-50-D8  | 512x512   |  320000 | -        | -              | 41.37 |         42.22 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3_r50-d8_512x512_4x4_320k_coco-stuff164k.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x512_4x4_320k_coco-stuff164k/deeplabv3_r50-d8_512x512_4x4_320k_coco-stuff164k_20210709_155403-51b21115.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x512_4x4_320k_coco-stuff164k/deeplabv3_r50-d8_512x512_4x4_320k_coco-stuff164k_20210709_155403.log.json)     |
 | DeepLabV3 | R-101-D8 | 512x512   |  320000 | -        | -              | 42.61 |         43.42 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k.py) | [model](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k_20210709_155402-3cbca14d.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k_20210709_155402.log.json) |
+
+Note:
+
+- `FP16` means Mixed Precision (FP16) is adopted in training.
diff --git a/configs/deeplabv3/deeplabv3.yml b/configs/deeplabv3/deeplabv3.yml
@@ -157,6 +157,27 @@ Models:
       mIoU(ms+flip): 81.21
   Config: configs/deeplabv3/deeplabv3_r101-d8_512x1024_80k_cityscapes.py
   Weights: https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x1024_80k_cityscapes/deeplabv3_r101-d8_512x1024_80k_cityscapes_20200606_113503-9e428899.pth
+- Name: deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes
+  In Collection: deeplabv3
+  Metadata:
+    backbone: R-101-D8
+    crop size: (512,1024)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 259.07
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP16
+      resolution: (512,1024)
+    memory (GB): 5.75
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: Cityscapes
+    Metrics:
+      mIoU: 80.48
+  Config: configs/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes/deeplabv3_r101-d8_fp16_512x1024_80k_cityscapes_20200717_230920-774d9cec.pth
 - Name: deeplabv3_r18-d8_769x769_80k_cityscapes
   In Collection: deeplabv3
   Metadata:

diff --git a/...t_r101-d8_512x1024_80k_fp16_cityscapes.py → ...3_r101-d8_fp16_512x1024_80k_cityscapes.py b/...t_r101-d8_512x1024_80k_fp16_cityscapes.py → ...3_r101-d8_fp16_512x1024_80k_cityscapes.py
@@ -1,4 +1,4 @@
-_base_ = '../pspnet/pspnet_r101-d8_512x1024_80k_cityscapes.py'
+_base_ = './deeplabv3_r101-d8_512x1024_80k_cityscapes.py'
 # fp16 settings
 optimizer_config = dict(type='Fp16OptimizerHook', loss_scale=512.)
 # fp16 placeholder