Add Pascal Context to mmsegmentation (open-mmlab#133)

* Add Pascal Context to mmsegmentation

* Add benchmark result to Pascal Context

* fix mmcv version

* fix code syntax

* fix code syntax again

* Update mmseg/models/segmentors/encoder_decoder.py

update hint

Co-authored-by: Jerry Jiarui XU <[email protected]>

* update comment

* fix pascal context model path

* fix model path mistake again

* fix model path mistake again

* fix model path mistakes again

Co-authored-by: Jerry Jiarui XU <[email protected]>

Author: yamengxi

configs/_base_/datasets/pascal_context.py

@@ -0,0 +1,60 @@
+# dataset settings
+dataset_type = 'PascalContextDataset'
+data_root = 'data/VOCdevkit/VOC2010/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+img_scale = (520, 520)
+crop_size = (480, 480)
+
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations'),
+    dict(type='Resize', img_scale=img_scale, ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
+    dict(type='RandomFlip', flip_ratio=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=img_scale,
+        # 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(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='JPEGImages',
+        ann_dir='SegmentationClassContext',
+        split='ImageSets/SegmentationContext/train.txt',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='JPEGImages',
+        ann_dir='SegmentationClassContext',
+        split='ImageSets/SegmentationContext/val.txt',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        img_dir='JPEGImages',
+        ann_dir='SegmentationClassContext',
+        split='ImageSets/SegmentationContext/val.txt',
+        pipeline=test_pipeline))

configs/deeplabv3/README.md

@@ -41,3 +41,9 @@ Note: `D-8` here corresponding to the output stride 8 setting for DeepLab series
 | DeepLabV3 | R-101-D8 | 512x512   |   20000 |      9.6 |           9.81 | 78.70 |         79.95 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_20k_voc12aug/deeplabv3_r101-d8_512x512_20k_voc12aug_20200617_010932-8d13832f.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_20k_voc12aug/deeplabv3_r101-d8_512x512_20k_voc12aug_20200617_010932.log.json) |
 | DeepLabV3 | R-50-D8  | 512x512   |   40000 | -        | -              | 77.68 |         78.78 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x512_40k_voc12aug/deeplabv3_r50-d8_512x512_40k_voc12aug_20200613_161546-2ae96e7e.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r50-d8_512x512_40k_voc12aug/deeplabv3_r50-d8_512x512_40k_voc12aug_20200613_161546.log.json)     |
 | DeepLabV3 | R-101-D8 | 512x512   |   40000 | -        | -              | 77.92 |         79.18 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_40k_voc12aug/deeplabv3_r101-d8_512x512_40k_voc12aug_20200613_161432-0017d784.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_40k_voc12aug/deeplabv3_r101-d8_512x512_40k_voc12aug_20200613_161432.log.json) |
+
+### Pascal Context
+|  Method   | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) |                                                                                                                                                                                                    download                                                                                                                                                                                                    |
+|-----------|----------|-----------|--------:|----------|----------------|------:|--------------:|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| DeepLabV3 | R-101-D8 | 480x480   |   40000 |      9.2 |           7.09 | 46.55 |         47.81 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_480x480_40k_pascal_context/deeplabv3_r101-d8_480x480_40k_pascal_context_20200911_204118-1aa27336.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_480x480_40k_pascal_context/deeplabv3_r101-d8_480x480_40k_pascal_context-20200911_204118.log.json) |
+| DeepLabV3 | R-101-D8 | 480x480   |   80000 | -        | -              | 46.42 |         47.53 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_480x480_80k_pascal_context/deeplabv3_r101-d8_480x480_80k_pascal_context_20200911_170155-2a21fff3.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_480x480_80k_pascal_context/deeplabv3_r101-d8_480x480_80k_pascal_context-20200911_170155.log.json) |

configs/deeplabv3/deeplabv3_r101-d8_480x480_40k_pascal_context.py

@@ -0,0 +1,2 @@
+_base_ = './deeplabv3_r50-d8_480x480_40k_pascal_context.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/deeplabv3/deeplabv3_r101-d8_480x480_80k_pascal_context.py

@@ -0,0 +1,2 @@
+_base_ = './deeplabv3_r50-d8_480x480_80k_pascal_context.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/deeplabv3/deeplabv3_r50-d8_480x480_40k_pascal_context.py

@@ -0,0 +1,9 @@
+_base_ = [
+    '../_base_/models/deeplabv3_r50-d8.py',
+    '../_base_/datasets/pascal_context.py', '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_40k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=60), auxiliary_head=dict(num_classes=60))
+test_cfg = dict(mode='slide', crop_size=(480, 480), stride=(320, 320))
+optimizer = dict(type='SGD', lr=0.004, momentum=0.9, weight_decay=0.0001)

configs/deeplabv3/deeplabv3_r50-d8_480x480_80k_pascal_context.py

@@ -0,0 +1,9 @@
+_base_ = [
+    '../_base_/models/deeplabv3_r50-d8.py',
+    '../_base_/datasets/pascal_context.py', '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_80k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=60), auxiliary_head=dict(num_classes=60))
+test_cfg = dict(mode='slide', crop_size=(480, 480), stride=(320, 320))
+optimizer = dict(type='SGD', lr=0.004, momentum=0.9, weight_decay=0.0001)

configs/deeplabv3plus/README.md

@@ -41,3 +41,9 @@ Note: `D-8` here corresponding to the output stride 8 setting for DeepLab series
 | DeepLabV3+ | R-101-D8 | 512x512   |   20000 |       11 |          13.88 | 77.22 |         78.59 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_512x512_20k_voc12aug/deeplabv3plus_r101-d8_512x512_20k_voc12aug_20200617_102345-c7ff3d56.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_512x512_20k_voc12aug/deeplabv3plus_r101-d8_512x512_20k_voc12aug_20200617_102345.log.json) |
 | DeepLabV3+ | R-50-D8  | 512x512   |   40000 | -        | -              | 76.81 |         77.57 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_512x512_40k_voc12aug/deeplabv3plus_r50-d8_512x512_40k_voc12aug_20200613_161759-e1b43aa9.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r50-d8_512x512_40k_voc12aug/deeplabv3plus_r50-d8_512x512_40k_voc12aug_20200613_161759.log.json)     |
 | DeepLabV3+ | R-101-D8 | 512x512   |   40000 | -        | -              | 78.62 |         79.53 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_512x512_40k_voc12aug/deeplabv3plus_r101-d8_512x512_40k_voc12aug_20200613_205333-faf03387.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_512x512_40k_voc12aug/deeplabv3plus_r101-d8_512x512_40k_voc12aug_20200613_205333.log.json) |
+
+#### Pascal Context
+|   Method   | Backbone | Crop Size | Lr schd | Mem (GB) | Inf time (fps) | mIoU  | mIoU(ms+flip) |                                                                                                                                                                                                                download                                                                                                                                                                                                                |
+|------------|----------|-----------|--------:|----------|----------------|------:|--------------:|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| DeepLabV3+ | R-101-D8 | 480x480   |   40000 |       - |          9.09 | 47.30 |         48.47 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_480x480_40k_pascal_context/deeplabv3plus_r101-d8_480x480_40k_pascal_context_20200911_165459-d3c8a29e.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_480x480_40k_pascal_context/deeplabv3plus_r101-d8_480x480_40k_pascal_context-20200911_165459.log.json) |
+| DeepLabV3+ | R-101-D8 | 480x480   |   80000 | -        | -              | 47.23 |         48.26 | [model](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_480x480_80k_pascal_context/deeplabv3plus_r101-d8_480x480_80k_pascal_context_20200911_155322-145d3ee8.pth) | [log](https://openmmlab.oss-accelerate.aliyuncs.com/mmsegmentation/v0.5/deeplabv3plus/deeplabv3plus_r101-d8_480x480_80k_pascal_context/deeplabv3plus_r101-d8_480x480_80k_pascal_context-20200911_155322.log.json) |

configs/deeplabv3plus/deeplabv3plus_r101-d8_480x480_40k_pascal_context.py

@@ -0,0 +1,2 @@
+_base_ = './deeplabv3plus_r50-d8_480x480_40k_pascal_context.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/deeplabv3plus/deeplabv3plus_r101-d8_480x480_80k_pascal_context.py

@@ -0,0 +1,2 @@
+_base_ = './deeplabv3plus_r50-d8_480x480_80k_pascal_context.py'
+model = dict(pretrained='open-mmlab://resnet101_v1c', backbone=dict(depth=101))

configs/deeplabv3plus/deeplabv3plus_r50-d8_480x480_40k_pascal_context.py

@@ -0,0 +1,9 @@
+_base_ = [
+    '../_base_/models/deeplabv3plus_r50-d8.py',
+    '../_base_/datasets/pascal_context.py', '../_base_/default_runtime.py',
+    '../_base_/schedules/schedule_40k.py'
+]
+model = dict(
+    decode_head=dict(num_classes=60), auxiliary_head=dict(num_classes=60))
+test_cfg = dict(mode='slide', crop_size=(480, 480), stride=(320, 320))
+optimizer = dict(type='SGD', lr=0.004, momentum=0.9, weight_decay=0.0001)