[Feature] add setr cityscapes benchmark (open-mmlab#1087)

RockeyCoss · Junjun2016 · web-flow · commit 29182206672d · 2021-12-03T00:18:50.000-08:00
* [Feature] add setr cityscapes benchmark

* change pretrain

* Update configs/_base_/datasets/cityscapes_768x768.py

Co-authored-by: Junjun2016 &lt;hejunjun@sjtu.edu.cn&gt;

* remove redundant keys

* remove redundant keys

* fix lint error

* update readme

* update pretrain

Co-authored-by: Junjun2016 &lt;hejunjun@sjtu.edu.cn&gt;
diff --git a/configs/_base_/datasets/cityscapes_768x768.py b/configs/_base_/datasets/cityscapes_768x768.py
@@ -0,0 +1,35 @@
+_base_ = './cityscapes.py'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (768, 768)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations'),
+    dict(type='Resize', img_scale=(2049, 1025), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2049, 1025),
+        # img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    train=dict(pipeline=train_pipeline),
+    val=dict(pipeline=test_pipeline),
+    test=dict(pipeline=test_pipeline))
diff --git a/configs/setr/README.md b/configs/setr/README.md
@@ -45,3 +45,11 @@ This head has two version head.
 | SETR-PUP | ViT-L | 512x512  | 16          | 160000   | 19.54        | 4.50              | 48.24 |             49.99 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_pup_512x512_160k_b16_ade20k.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_pup_512x512_160k_b16_ade20k/setr_pup_512x512_160k_b16_ade20k_20210619_191343-7e0ce826.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_pup_512x512_160k_b16_ade20k/setr_pup_512x512_160k_b16_ade20k_20210619_191343.log.json)     |
 | SETR-MLA | ViT-L | 512x512  | 8           | 160000   | 10.96        | -              | 47.34 |             49.05 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_mla_512x512_160k_b8_ade20k.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_512x512_160k_b8_ade20k/setr_mla_512x512_160k_b8_ade20k_20210619_191118-c6d21df0.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_512x512_160k_b8_ade20k/setr_mla_512x512_160k_b8_ade20k_20210619_191118.log.json)     |
 | SETR-MLA | ViT-L | 512x512  | 16          | 160000   | 17.30        | 5.25              | 47.54 |             49.37 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_mla_512x512_160k_b16_ade20k.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_512x512_160k_b16_ade20k/setr_mla_512x512_160k_b16_ade20k_20210619_191057-f9741de7.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_512x512_160k_b16_ade20k/setr_mla_512x512_160k_b16_ade20k_20210619_191057.log.json)     |
+
+### Cityscapes
+
+| Method | Backbone | Crop Size | Batch Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) | config                                                                                                                          | download                                                                                                                                                                                                                                                                                                                                                     |
+| ------ | -------- | --------- | ---------- | ------- | -------- | -------------- | ----- | ------------: | ------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| SETR-Naive | ViT-L | 768x768 | 8         | 80000 | 24.06 | 0.39 | 78.10 | 80.22 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_vit-large_naive_8x1_768x768_80k_cityscapes.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_naive_vit-large_8x1_768x768_80k_cityscapes/setr_naive_vit-large_8x1_768x768_80k_cityscapes_20211123_000505-20728e80.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_naive_vit-large_8x1_768x768_80k_cityscapes/setr_naive_vit-large_8x1_768x768_80k_cityscapes_20211123_000505.log.json)     |
+| SETR-PUP | ViT-L | 768x768 | 8         | 80000   | 27.96 | 0.37 | 79.21 | 81.02 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_vit-large_pup_8x1_768x768_80k_cityscapes.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_pup_vit-large_8x1_768x768_80k_cityscapes/setr_pup_vit-large_8x1_768x768_80k_cityscapes_20211122_155115-f6f37b8f.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_pup_vit-large_8x1_768x768_80k_cityscapes/setr_pup_vit-large_8x1_768x768_80k_cityscapes_20211122_155115.log.json)     |
+| SETR-MLA | ViT-L | 768x768 | 8           | 80000 | 24.10 | 0.41 | 77.00 | 79.59 | [config](https://github.com/open-mmlab/mmsegmentation/blob/master/configs/setr/setr_vit-large_mla_8x1_768x768_80k_cityscapes.py)  | [model](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_vit-large_8x1_768x768_80k_cityscapes/setr_mla_vit-large_8x1_768x768_80k_cityscapes_20211119_101003-7f8dccbe.pth) &#124; [log](https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_vit-large_8x1_768x768_80k_cityscapes/setr_mla_vit-large_8x1_768x768_80k_cityscapes_20211119_101003.log.json)     |
diff --git a/configs/setr/setr.yml b/configs/setr/setr.yml
@@ -3,6 +3,7 @@ Collections:
   Metadata:
     Training Data:
     - ADE20K
+    - Cityscapes
   Paper:
     URL: https://arxiv.org/abs/2012.15840
     Title: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective
@@ -95,3 +96,69 @@ Models:
       mIoU(ms+flip): 49.37
   Config: configs/setr/setr_mla_512x512_160k_b16_ade20k.py
   Weights: https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_512x512_160k_b16_ade20k/setr_mla_512x512_160k_b16_ade20k_20210619_191057-f9741de7.pth
+- Name: setr_vit-large_naive_8x1_768x768_80k_cityscapes
+  In Collection: setr
+  Metadata:
+    backbone: ViT-L
+    crop size: (768,768)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 2564.1
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (768,768)
+    Training Memory (GB): 24.06
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: Cityscapes
+    Metrics:
+      mIoU: 78.1
+      mIoU(ms+flip): 80.22
+  Config: configs/setr/setr_vit-large_naive_8x1_768x768_80k_cityscapes.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_naive_vit-large_8x1_768x768_80k_cityscapes/setr_naive_vit-large_8x1_768x768_80k_cityscapes_20211123_000505-20728e80.pth
+- Name: setr_vit-large_pup_8x1_768x768_80k_cityscapes
+  In Collection: setr
+  Metadata:
+    backbone: ViT-L
+    crop size: (768,768)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 2702.7
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (768,768)
+    Training Memory (GB): 27.96
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: Cityscapes
+    Metrics:
+      mIoU: 79.21
+      mIoU(ms+flip): 81.02
+  Config: configs/setr/setr_vit-large_pup_8x1_768x768_80k_cityscapes.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_pup_vit-large_8x1_768x768_80k_cityscapes/setr_pup_vit-large_8x1_768x768_80k_cityscapes_20211122_155115-f6f37b8f.pth
+- Name: setr_vit-large_mla_8x1_768x768_80k_cityscapes
+  In Collection: setr
+  Metadata:
+    backbone: ViT-L
+    crop size: (768,768)
+    lr schd: 80000
+    inference time (ms/im):
+    - value: 2439.02
+      hardware: V100
+      backend: PyTorch
+      batch size: 1
+      mode: FP32
+      resolution: (768,768)
+    Training Memory (GB): 24.1
+  Results:
+  - Task: Semantic Segmentation
+    Dataset: Cityscapes
+    Metrics:
+      mIoU: 77.0
+      mIoU(ms+flip): 79.59
+  Config: configs/setr/setr_vit-large_mla_8x1_768x768_80k_cityscapes.py
+  Weights: https://download.openmmlab.com/mmsegmentation/v0.5/setr/setr_mla_vit-large_8x1_768x768_80k_cityscapes/setr_mla_vit-large_8x1_768x768_80k_cityscapes_20211119_101003-7f8dccbe.pth
diff --git a/configs/setr/setr_mla_512x512_160k_b8_ade20k.py b/configs/setr/setr_mla_512x512_160k_b8_ade20k.py
@@ -4,8 +4,11 @@
 ]
 norm_cfg = dict(type='SyncBN', requires_grad=True)
 model = dict(
-    pretrained='pretrain/vit_large_patch16_384.pth',
-    backbone=dict(img_size=(512, 512), drop_rate=0.),
+    pretrained=None,
+    backbone=dict(
+        img_size=(512, 512),
+        drop_rate=0.,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
     decode_head=dict(num_classes=150),
     auxiliary_head=[
         dict(
diff --git a/configs/setr/setr_naive_512x512_160k_b16_ade20k.py b/configs/setr/setr_naive_512x512_160k_b16_ade20k.py
@@ -4,8 +4,11 @@
 ]
 norm_cfg = dict(type='SyncBN', requires_grad=True)
 model = dict(
-    pretrained='pretrain/vit_large_patch16_384.pth',
-    backbone=dict(img_size=(512, 512), drop_rate=0.),
+    pretrained=None,
+    backbone=dict(
+        img_size=(512, 512),
+        drop_rate=0.,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
     decode_head=dict(num_classes=150),
     auxiliary_head=[
         dict(
diff --git a/configs/setr/setr_pup_512x512_160k_b16_ade20k.py b/configs/setr/setr_pup_512x512_160k_b16_ade20k.py
@@ -4,8 +4,11 @@
 ]
 norm_cfg = dict(type='SyncBN', requires_grad=True)
 model = dict(
-    pretrained='pretrain/vit_large_patch16_384.pth',
-    backbone=dict(img_size=(512, 512), drop_rate=0.),
+    pretrained=None,
+    backbone=dict(
+        img_size=(512, 512),
+        drop_rate=0.,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
     decode_head=dict(num_classes=150),
     auxiliary_head=[
         dict(
diff --git a/configs/setr/setr_vit-large_mla_8x1_768x768_80k_cityscapes.py b/configs/setr/setr_vit-large_mla_8x1_768x768_80k_cityscapes.py
@@ -0,0 +1,16 @@
+_base_ = [
+    '../_base_/models/setr_mla.py', '../_base_/datasets/cityscapes_768x768.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    pretrained=None,
+    backbone=dict(
+        drop_rate=0,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
+    test_cfg=dict(mode='slide', crop_size=(768, 768), stride=(512, 512)))
+
+optimizer = dict(
+    lr=0.002,
+    weight_decay=0.0,
+    paramwise_cfg=dict(custom_keys={'head': dict(lr_mult=10.)}))
+data = dict(samples_per_gpu=1)
diff --git a/configs/setr/setr_vit-large_naive_8x1_768x768_80k_cityscapes.py b/configs/setr/setr_vit-large_naive_8x1_768x768_80k_cityscapes.py
@@ -0,0 +1,17 @@
+_base_ = [
+    '../_base_/models/setr_naive.py',
+    '../_base_/datasets/cityscapes_768x768.py', '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    pretrained=None,
+    backbone=dict(
+        drop_rate=0.,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
+    test_cfg=dict(mode='slide', crop_size=(768, 768), stride=(512, 512)))
+
+optimizer = dict(
+    weight_decay=0.0,
+    paramwise_cfg=dict(custom_keys={'head': dict(lr_mult=10.)}))
+
+data = dict(samples_per_gpu=1)
diff --git a/configs/setr/setr_vit-large_pup_8x1_768x768_80k_cityscapes.py b/configs/setr/setr_vit-large_pup_8x1_768x768_80k_cityscapes.py
@@ -0,0 +1,63 @@
+_base_ = [
+    '../_base_/models/setr_pup.py', '../_base_/datasets/cityscapes_768x768.py',
+    '../_base_/default_runtime.py', '../_base_/schedules/schedule_80k.py'
+]
+
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+crop_size = (768, 768)
+model = dict(
+    pretrained=None,
+    backbone=dict(
+        drop_rate=0.,
+        init_cfg=dict(type='Pretrained', checkpoint='mmcls://vit_large_p16')),
+    auxiliary_head=[
+        dict(
+            type='SETRUPHead',
+            in_channels=1024,
+            channels=256,
+            in_index=0,
+            num_classes=19,
+            dropout_ratio=0,
+            norm_cfg=norm_cfg,
+            num_convs=2,
+            up_scale=4,
+            kernel_size=3,
+            align_corners=False,
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)),
+        dict(
+            type='SETRUPHead',
+            in_channels=1024,
+            channels=256,
+            in_index=1,
+            num_classes=19,
+            dropout_ratio=0,
+            norm_cfg=norm_cfg,
+            num_convs=2,
+            up_scale=4,
+            kernel_size=3,
+            align_corners=False,
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4)),
+        dict(
+            type='SETRUPHead',
+            in_channels=1024,
+            channels=256,
+            in_index=2,
+            num_classes=19,
+            dropout_ratio=0,
+            norm_cfg=norm_cfg,
+            num_convs=2,
+            up_scale=4,
+            kernel_size=3,
+            align_corners=False,
+            loss_decode=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.4))
+    ],
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(512, 512)))
+
+optimizer = dict(
+    weight_decay=0.0,
+    paramwise_cfg=dict(custom_keys={'head': dict(lr_mult=10.)}))
+
+data = dict(samples_per_gpu=1)
