add rotated rt detr

Author: wokaikaixinxin

projects/rotated_rtdetr/configs/default_runtime.py

@@ -0,0 +1,34 @@
+default_scope = 'ai4rs'
+
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=50),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', interval=1, max_keep_ckpts=99999),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='mmdet.DetVisualizationHook'))
+
+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='RotLocalVisualizer', vis_backends=vis_backends, name='visualizer')
+log_processor = dict(type='LogProcessor', window_size=50, by_epoch=True)
+
+log_level = 'INFO'
+load_from = None
+resume = False
+
+custom_hooks = [
+    dict(type='mmdet.NumClassCheckHook'),
+    dict(
+        type='EMAHook',
+        ema_type='mmdet.ExpMomentumEMA',
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49)
+]

projects/rotated_rtdetr/configs/dota_rr.py

@@ -0,0 +1,101 @@
+# dataset settings
+dataset_type = 'DOTADataset'
+data_root = 'data/split_ss_dota/'
+
+backend_args = None
+
+train_pipeline = [
+    dict(type='mmdet.LoadImageFromFile', backend_args=backend_args),
+    dict(type='mmdet.LoadAnnotations', with_bbox=True, box_type='qbox'),
+    dict(type='ConvertBoxType', box_type_mapping=dict(gt_bboxes='rbox')),
+    dict(type='mmdet.Resize', scale=(1024, 1024), keep_ratio=True),
+    dict(
+        type='mmdet.RandomFlip',
+        prob=0.75,
+        direction=['horizontal', 'vertical', 'diagonal']),
+    dict(
+        type='RandomRotate',
+        prob=0.5,
+        angle_range=180,
+        rect_obj_labels=[9, 11]),
+    dict(
+        type='mmdet.Pad', size=(1024, 1024),
+        pad_val=dict(img=(114, 114, 114))),
+    dict(type='mmdet.PackDetInputs')
+]
+val_pipeline = [
+    dict(type='mmdet.LoadImageFromFile', backend_args=backend_args),
+    dict(type='mmdet.Resize', scale=(1024, 1024), keep_ratio=True),
+    # avoid bboxes being resized
+    dict(type='mmdet.LoadAnnotations', with_bbox=True, box_type='qbox'),
+    dict(type='ConvertBoxType', box_type_mapping=dict(gt_bboxes='rbox')),
+    dict(
+        type='mmdet.Pad', size=(1024, 1024),
+        pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type='mmdet.PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor'))
+]
+test_pipeline = [
+    dict(type='mmdet.LoadImageFromFile', backend_args=backend_args),
+    dict(type='mmdet.Resize', scale=(1024, 1024), keep_ratio=True),
+    dict(
+        type='mmdet.Pad', size=(1024, 1024),
+        pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type='mmdet.PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor'))
+]
+train_dataloader = dict(
+    batch_size=4,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type='DefaultSampler', shuffle=True),
+    batch_sampler=None,
+    pin_memory=False,
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file='trainval/annfiles/',
+        data_prefix=dict(img_path='trainval/images/'),
+        filter_cfg=dict(filter_empty_gt=True),
+        pipeline=train_pipeline))
+val_dataloader = dict(
+    batch_size=4,
+    num_workers=4,
+    persistent_workers=True,
+    drop_last=False,
+    sampler=dict(type='DefaultSampler', shuffle=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file='trainval/annfiles/',
+        data_prefix=dict(img_path='trainval/images/'),
+        test_mode=True,
+        pipeline=val_pipeline))
+# test_dataloader = val_dataloader
+
+val_evaluator = dict(type='DOTAMetric', metric='mAP')
+# test_evaluator = val_evaluator
+
+# inference on test dataset and format the output results
+# for submission. Note: the test set has no annotation.
+test_dataloader = dict(
+    batch_size=4,
+    num_workers=4,
+    persistent_workers=False,
+    drop_last=False,
+    sampler=dict(type='DefaultSampler', shuffle=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_prefix=dict(img_path='test/images/'),
+        test_mode=True,
+        pipeline=test_pipeline))
+test_evaluator = dict(
+    type='DOTAMetric',
+    format_only=True,
+    merge_patches=True,
+    outfile_prefix='./work_dirs/Task1')

projects/rotated_rtdetr/configs/o2_rtdetr_r101vd_2xb4_72e_dota.py

@@ -0,0 +1,22 @@
+from mmengine.config import read_base
+with read_base():
+    from .o2_rtdetr_r50vd_2xb4_72e_dota import *
+
+pretrained = ('https://www.modelscope.cn/models/wokaikaixinxin/ai4rs/resolve/'
+              'master/rtdetr/resnet101vd_ssld_pretrained_64ed664a.pth')  # noqa
+
+model.update(
+    backbone=dict(
+        depth=101, init_cfg=dict(type='Pretrained', checkpoint=pretrained)),
+    neck=dict(out_channels=384),
+    encoder=dict(
+        in_channels=[384, 384, 384],
+        fpn_cfg=dict(in_channels=[384, 384, 384]),
+        layer_cfg=dict(
+            self_attn_cfg=dict(embed_dims=384),
+            ffn_cfg=dict(embed_dims=384, feedforward_channels=2048))))
+
+# optimizer
+optim_wrapper.update(
+    paramwise_cfg=dict(
+        custom_keys={'backbone': dict(lr_mult=0.01)}, norm_decay_mult=1))

projects/rotated_rtdetr/configs/o2_rtdetr_r18vd_2xb4_72e_dota.py

@@ -0,0 +1,41 @@
+from mmengine.config import read_base
+with read_base():
+    from .o2_rtdetr_r50vd_2xb4_72e_dota import *
+
+pretrained = ('https://www.modelscope.cn/models/wokaikaixinxin/ai4rs/resolve/'
+              'master/rtdetr/resnet18vd_pretrained_55f5a0d6.pth')  # noqa
+
+model.update(
+    backbone=dict(
+        depth=18,
+        frozen_stages=-1,
+        norm_cfg=dict(requires_grad=True),
+        norm_eval=False,
+        init_cfg=dict(type='Pretrained', checkpoint=pretrained)),
+    neck=dict(in_channels=[128, 256, 512]),
+    encoder=dict(fpn_cfg=dict(expansion=0.5)),
+    decoder=dict(num_layers=3))
+
+# set all norm layers in backbone to lr_mult=0.1 and decay_mult=0.0
+# set all other layers in backbone to lr_mult=0.1
+num_blocks_list = (2, 2, 2, 2)  # r18
+downsample_norm_idx_list = (2, 3, 3, 3)  # r18
+backbone_norm_multi = dict(lr_mult=0.1, decay_mult=0.0)
+custom_keys = {'backbone': dict(lr_mult=0.1)}
+custom_keys.update({
+    f'backbone.layer{stage_id + 1}.{block_id}.bn': backbone_norm_multi
+    for stage_id, num_blocks in enumerate(num_blocks_list)
+    for block_id in range(num_blocks)
+})
+custom_keys.update({
+    f'backbone.layer{stage_id + 1}.{block_id}.downsample.{downsample_norm_idx - 1}':  # noqa
+    backbone_norm_multi
+    for stage_id, (num_blocks, downsample_norm_idx) in enumerate(
+        zip(num_blocks_list, downsample_norm_idx_list))
+    for block_id in range(num_blocks)
+})
+
+# optimizer
+optim_wrapper.paramwise_cfg.pop('custom_keys')
+optim_wrapper.update(
+    paramwise_cfg=dict(custom_keys=dict(**custom_keys)))

projects/rotated_rtdetr/configs/o2_rtdetr_r34vd_2xb4_72e_dota.py

@@ -0,0 +1,41 @@
+from mmengine.config import read_base
+with read_base():
+    from .o2_rtdetr_r50vd_2xb4_72e_dota import *
+
+pretrained = ('https://www.modelscope.cn/models/wokaikaixinxin/ai4rs/resolve/'
+              'master/rtdetr/resnet34vd_pretrained_f6a72dc5.pth')  # noqa
+
+model.update(
+    backbone=dict(
+        depth=34,
+        frozen_stages=-1,
+        norm_cfg=dict(requires_grad=True),
+        norm_eval=False,
+        init_cfg=dict(type='Pretrained', checkpoint=pretrained)),
+    neck=dict(in_channels=[128, 256, 512]),
+    encoder=dict(fpn_cfg=dict(expansion=0.5)),
+    decoder=dict(num_layers=4))
+
+# set all norm layers in backbone to lr_mult=0.1 and decay_mult=0.0
+# set all other layers in backbone to lr_mult=0.1
+num_blocks_list = (3, 4, 6, 3)  # r34
+downsample_norm_idx_list = (2, 3, 3, 3)  # r34
+backbone_norm_multi = dict(lr_mult=0.1, decay_mult=0.0)
+custom_keys = {'backbone': dict(lr_mult=0.1)}
+custom_keys.update({
+    f'backbone.layer{stage_id + 1}.{block_id}.bn': backbone_norm_multi
+    for stage_id, num_blocks in enumerate(num_blocks_list)
+    for block_id in range(num_blocks)
+})
+custom_keys.update({
+    f'backbone.layer{stage_id + 1}.{block_id}.downsample.{downsample_norm_idx - 1}':  # noqa
+    backbone_norm_multi
+    for stage_id, (num_blocks, downsample_norm_idx) in enumerate(
+        zip(num_blocks_list, downsample_norm_idx_list))
+    for block_id in range(num_blocks)
+})
+
+# optimizer
+optim_wrapper.paramwise_cfg.pop('custom_keys')
+optim_wrapper.update(
+    paramwise_cfg=dict(custom_keys=dict(**custom_keys)))