[New configs]Segformer_ade20k

mmseg/configs/_base_/datasets/ade20k.py

@@ -0,0 +1,77 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmcv.transforms import RandomFlip, RandomResize, Resize, TestTimeAug
+from mmcv.transforms.loading import LoadImageFromFile
+from mmengine.dataset.sampler import DefaultSampler, InfiniteSampler
+
+from mmseg.datasets.ade import ADE20KDataset
+from mmseg.datasets.transforms import (LoadAnnotations, PackSegInputs,
+                                       PhotoMetricDistortion, RandomCrop)
+from mmseg.evaluation import IoUMetric
+
+# dataset settings
+dataset_type = ADE20KDataset
+data_root = 'data/ade/ADEChallengeData2016'
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=LoadAnnotations),
+    dict(
+        type=RandomResize,
+        scale=(2048, 512),
+        ratio_range=(0.5, 2.0),
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=crop_size, cat_max_ratio=0.75),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=PhotoMetricDistortion),
+    dict(type=PackSegInputs)
+]
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=Resize, scale=(2048, 512), keep_ratio=True),
+    # add loading annotation after ``Resize`` because ground truth
+    # does not need to do resize data transform
+    dict(type=LoadAnnotations),
+    dict(type=PackSegInputs)
+]
+img_ratios = [0.5, 0.75, 1.0, 1.25, 1.5, 1.75]
+tta_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=None),
+    dict(
+        type=TestTimeAug,
+        transforms=[[
+            dict(type=Resize, scale_factor=r, keep_ratio=True)
+            for r in img_ratios
+        ],
+                    [
+                        dict(type=RandomFlip, prob=0., direction='horizontal'),
+                        dict(type=RandomFlip, prob=1., direction='horizontal')
+                    ], [dict(type=LoadAnnotations)],
+                    [dict(type=PackSegInputs)]])
+]
+train_dataloader = dict(
+    batch_size=4,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type=InfiniteSampler, shuffle=True),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_prefix=dict(
+            img_path='images/training', seg_map_path='annotations/training'),
+        pipeline=train_pipeline))
+val_dataloader = dict(
+    batch_size=1,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type=DefaultSampler, shuffle=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_prefix=dict(
+            img_path='images/validation',
+            seg_map_path='annotations/validation'),
+        pipeline=test_pipeline))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(type=IoUMetric, iou_metrics=['mIoU'])
+test_evaluator = val_evaluator

mmseg/configs/_base_/datasets/ade20k_640x640.py

@@ -0,0 +1,77 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmcv.transforms import RandomFlip, RandomResize, Resize, TestTimeAug
+from mmcv.transforms.loading import LoadImageFromFile
+from mmengine.dataset.sampler import DefaultSampler, InfiniteSampler
+
+from mmseg.datasets.ade import ADE20KDataset
+from mmseg.datasets.transforms import (LoadAnnotations, PackSegInputs,
+                                       PhotoMetricDistortion, RandomCrop)
+from mmseg.evaluation import IoUMetric
+
+# dataset settings
+dataset_type = ADE20KDataset
+data_root = 'data/ade/ADEChallengeData2016'
+crop_size = (640, 640)
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=LoadAnnotations),
+    dict(
+        type=RandomResize,
+        scale=(2560, 640),
+        ratio_range=(0.5, 2.0),
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=crop_size, cat_max_ratio=0.75),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=PhotoMetricDistortion),
+    dict(type=PackSegInputs)
+]
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=Resize, scale=(2560, 640), keep_ratio=True),
+    # add loading annotation after ``Resize`` because ground truth
+    # does not need to do resize data transform
+    dict(type=LoadAnnotations),
+    dict(type=PackSegInputs)
+]
+img_ratios = [0.5, 0.75, 1.0, 1.25, 1.5, 1.75]
+tta_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=None),
+    dict(
+        type=TestTimeAug,
+        transforms=[[
+            dict(type=Resize, scale_factor=r, keep_ratio=True)
+            for r in img_ratios
+        ],
+                    [
+                        dict(type=RandomFlip, prob=0., direction='horizontal'),
+                        dict(type=RandomFlip, prob=1., direction='horizontal')
+                    ], [dict(type=LoadAnnotations)],
+                    [dict(type=PackSegInputs)]])
+]
+train_dataloader = dict(
+    batch_size=4,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type=InfiniteSampler, shuffle=True),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_prefix=dict(
+            img_path='images/training', seg_map_path='annotations/training'),
+        pipeline=train_pipeline))
+val_dataloader = dict(
+    batch_size=1,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type=DefaultSampler, shuffle=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_prefix=dict(
+            img_path='images/validation',
+            seg_map_path='annotations/validation'),
+        pipeline=test_pipeline))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(type=IoUMetric, iou_metrics=['mIoU'])
+test_evaluator = val_evaluator

mmseg/configs/_base_/default_runtime.py

@@ -0,0 +1,24 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.runner import LogProcessor
+from mmengine.visualization import LocalVisBackend
+
+from mmseg.models.segmentors import SegTTAModel
+from mmseg.visualization import SegLocalVisualizer
+
+default_scope = None
+env_cfg = dict(
+    cudnn_benchmark=False,
+    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
+    dist_cfg=dict(backend='nccl'),
+)
+
+vis_backends = [dict(type=LocalVisBackend)]
+visualizer = dict(
+    type=SegLocalVisualizer, vis_backends=vis_backends, name='visualizer')
+log_processor = dict(type=LogProcessor, window_size=50, by_epoch=False)
+
+log_level = 'INFO'
+load_from = None
+resume = False
+
+tta_model = dict(type=SegTTAModel)

mmseg/configs/_base_/models/segformer_mit_b0.py

@@ -0,0 +1,49 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmseg.models.backbones import MixVisionTransformer
+from mmseg.models.data_preprocessor import SegDataPreProcessor
+from mmseg.models.decode_heads import SegformerHead
+from mmseg.models.losses import CrossEntropyLoss
+from mmseg.models.segmentors import EncoderDecoder
+
+# model settings
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+data_preprocessor = dict(
+    type=SegDataPreProcessor,
+    mean=[123.675, 116.28, 103.53],
+    std=[58.395, 57.12, 57.375],
+    bgr_to_rgb=True,
+    pad_val=0,
+    seg_pad_val=255)
+model = dict(
+    type=EncoderDecoder,
+    data_preprocessor=data_preprocessor,
+    pretrained=None,
+    backbone=dict(
+        type=MixVisionTransformer,
+        in_channels=3,
+        embed_dims=32,
+        num_stages=4,
+        num_layers=[2, 2, 2, 2],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type=SegformerHead,
+        in_channels=[32, 64, 160, 256],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=19,
+        norm_cfg=norm_cfg,
+        align_corners=False,
+        loss_decode=dict(
+            type=CrossEntropyLoss, use_sigmoid=False, loss_weight=1.0)),
+    # model training and testing settings
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))

mmseg/configs/_base_/schedules/schedule_160k.py

@@ -0,0 +1,35 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
+                            LoggerHook, ParamSchedulerHook)
+from mmengine.optim.optimizer.optimizer_wrapper import OptimWrapper
+from mmengine.optim.scheduler.lr_scheduler import PolyLR
+from mmengine.runner.loops import IterBasedTrainLoop, TestLoop, ValLoop
+from torch.optim.sgd import SGD
+
+from mmseg.engine.hooks import SegVisualizationHook
+
+# optimizer
+optimizer = dict(type=SGD, lr=0.01, momentum=0.9, weight_decay=0.0005)
+optim_wrapper = dict(type=OptimWrapper, optimizer=optimizer, clip_grad=None)
+# learning policy
+param_scheduler = [
+    dict(
+        type=PolyLR,
+        eta_min=1e-4,
+        power=0.9,
+        begin=0,
+        end=160000,
+        by_epoch=False)
+]
+# training schedule for 160k
+train_cfg = dict(type=IterBasedTrainLoop, max_iters=160000, val_interval=16000)
+val_cfg = dict(type=ValLoop)
+test_cfg = dict(type=TestLoop)
+
+default_hooks = dict(
+    timer=dict(type=IterTimerHook),
+    logger=dict(type=LoggerHook, interval=50, log_metric_by_epoch=False),
+    param_scheduler=dict(type=ParamSchedulerHook),
+    checkpoint=dict(type=CheckpointHook, by_epoch=False, interval=16000),
+    sampler_seed=dict(type=DistSamplerSeedHook),
+    visualization=dict(type=SegVisualizationHook))

mmseg/configs/segformer/segformer_mit_b0_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,47 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.models.segformer_mit_b0 import *  # noqa: F401,F403
+    from .._base_.datasets.ade20k import *  # noqa: F401,F403
+    from .._base_.schedules.schedule_160k import *  # noqa: F401,F403
+    from .._base_.default_runtime import *  # noqa: F401,F403
+
+from mmengine.model.weight_init import PretrainedInit
+from mmengine.optim.optimizer.optimizer_wrapper import OptimWrapper
+from mmengine.optim.scheduler.lr_scheduler import LinearLR, PolyLR
+from torch.optim.adamw import AdamW
+
+crop_size = (512, 512)
+data_preprocessor.update(size=crop_size)  # noqa: F405
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+model.update(  # noqa: F405
+    data_preprocessor=data_preprocessor,  # noqa: F405
+    backbone=dict(init_cfg=dict(type=PretrainedInit, checkpoint=checkpoint)),
+    decode_head=dict(num_classes=150))
+
+optim_wrapper = dict(
+    type=OptimWrapper,
+    optimizer=dict(
+        type=AdamW, lr=0.00006, betas=(0.9, 0.999), weight_decay=0.01),
+    paramwise_cfg=dict(
+        custom_keys={
+            'pos_block': dict(decay_mult=0.),
+            'norm': dict(decay_mult=0.),
+            'head': dict(lr_mult=10.)
+        }))
+
+param_scheduler = [
+    dict(type=LinearLR, start_factor=1e-6, by_epoch=False, begin=0, end=1500),
+    dict(
+        type=PolyLR,
+        eta_min=0.0,
+        power=1.0,
+        begin=1500,
+        end=160000,
+        by_epoch=False,
+    )
+]
+train_dataloader.update(batch_size=2, num_workers=2)  # noqa: F405
+val_dataloader.update(batch_size=1, num_workers=4)  # noqa: F405
+test_dataloader = val_dataloader  # noqa: F405

mmseg/configs/segformer/segformer_mit_b1_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,16 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .segformer_mit_b0_8xb2_160k_ade20k_512x512 import *  # noqa: F401,F403
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b1_20220624-02e5a6a1.pth'  # noqa
+
+# model settings
+model.update(  # noqa: F405
+    backbone=dict(
+        init_cfg=dict(checkpoint=checkpoint),
+        embed_dims=64,
+        num_heads=[1, 2, 5, 8],
+        num_layers=[2, 2, 2, 2]),
+    decode_head=dict(in_channels=[64, 128, 320, 512]))

mmseg/configs/segformer/segformer_mit_b2_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,16 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .segformer_mit_b0_8xb2_160k_ade20k_512x512 import *  # noqa: F401,F403
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model.update(  # noqa: F405
+    backbone=dict(
+        init_cfg=dict(checkpoint=checkpoint),
+        embed_dims=64,
+        num_heads=[1, 2, 5, 8],
+        num_layers=[3, 4, 6, 3]),
+    decode_head=dict(in_channels=[64, 128, 320, 512]))

mmseg/configs/segformer/segformer_mit_b3_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,16 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .segformer_mit_b0_8xb2_160k_ade20k_512x512 import *  # noqa: F401,F403
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b3_20220624-13b1141c.pth'  # noqa
+
+# model settings
+model.update(  # noqa: F405
+    backbone=dict(
+        init_cfg=dict(checkpoint=checkpoint),
+        embed_dims=64,
+        num_heads=[1, 2, 5, 8],
+        num_layers=[3, 4, 18, 3]),
+    decode_head=dict(in_channels=[64, 128, 320, 512]))

mmseg/configs/segformer/segformer_mit_b4_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,16 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .segformer_mit_b0_8xb2_160k_ade20k_512x512 import *  # noqa: F401,F403
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b4_20220624-d588d980.pth'  # noqa
+
+# model settings
+model.update(  # noqa: F405
+    backbone=dict(
+        init_cfg=dict(checkpoint=checkpoint),
+        embed_dims=64,
+        num_heads=[1, 2, 5, 8],
+        num_layers=[3, 8, 27, 3]),
+    decode_head=dict(in_channels=[64, 128, 320, 512]))

mmseg/configs/segformer/segformer_mit_b5_8xb2_160k_ade20k_512x512.py

@@ -0,0 +1,16 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .segformer_mit_b0_8xb2_160k_ade20k_512x512 import *  # noqa: F401,F403
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b5_20220624-658746d9.pth'  # noqa
+
+# model settings
+model.update(  # noqa: F405
+    backbone=dict(
+        init_cfg=dict(checkpoint=checkpoint),
+        embed_dims=64,
+        num_heads=[1, 2, 5, 8],
+        num_layers=[3, 6, 40, 3]),
+    decode_head=dict(in_channels=[64, 128, 320, 512]))