Code for "Grid R-CNN" (open-mmlab#810)

* Grid R-CNN

* add grid_rcnn_res50fpn2x config

* add assertion that grid_head should exist

* fix bugs and remove SharedFCBBoxHeadGrid

* remove the property with_grid

* format fixes for grad_head and add config dir

* move random_jitter to grid_head and some refactoring

* simplify the calculation of num_edges

* refactoring

* refactoring

* rename config files and add x101 config

* bug fix for inference

* remove random_jitter_single

* add readme of grid rcnn

* add bibtex of grid rcnn plus

* update work_dir

Author: luxiin

configs/grid_rcnn/README.md

@@ -0,0 +1,33 @@
+# Grid R-CNN
+
+## Introduction
+
+```
+@inproceedings{lu2019grid,
+  title={Grid r-cnn},
+  author={Lu, Xin and Li, Buyu and Yue, Yuxin and Li, Quanquan and Yan, Junjie},
+  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
+  year={2019}
+}
+
+@article{lu2019grid,
+  title={Grid R-CNN Plus: Faster and Better},
+  author={Lu, Xin and Li, Buyu and Yue, Yuxin and Li, Quanquan and Yan, Junjie},
+  journal={arXiv preprint arXiv:1906.05688},
+  year={2019}
+}
+```
+
+## Results and Models
+
+| Backbone    | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | Download |
+|:-----------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:--------:|
+| R-50        | 2x      | 4.8      |                     |                | 40.3   | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/grid_rcnn/grid_rcnn_gn_head_r50_fpn_2x_20190619-5b29cf9d.pth) |
+| R-101       | 2x      | 6.7      |                     |                | 41.7   | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/grid_rcnn/grid_rcnn_gn_head_r101_fpn_2x_20190619-a4b61645.pth) |
+| X-101-32x4d | 2x      | 8.0      |                     |                | 43.0   | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/grid_rcnn/grid_rcnn_gn_head_x101_32x4d_fpn_2x_20190619-0bbfd87a.pth) |
+| X-101-64x4d | 2x      | 10.9     |                     |                | 43.1   | [model](https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection/models/grid_rcnn/grid_rcnn_gn_head_x101_64x4d_fpn_2x_20190619-8f4e20bb.pth) |
+
+**Notes:**
+- All models are trained with 8 GPUs instead of 32 GPUs in the original paper.
+- The warming up lasts for 1 epoch and `2x` here indicates 25 epochs.
+- The training speed is about 3 times slower than Faster R-CNN.

configs/grid_rcnn/grid_rcnn_gn_head_r50_fpn_2x.py

@@ -0,0 +1,178 @@
+# model settings
+model = dict(
+    type='GridRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch'),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        loss_cls=dict(
+            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+        loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        with_reg=False,
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False),
+    grid_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    grid_head=dict(
+        type='GridHead',
+        grid_points=9,
+        num_convs=8,
+        in_channels=256,
+        point_feat_channels=64,
+        norm_cfg=dict(type='GN', num_groups=36),
+        loss_grid=dict(
+            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=15)))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        debug=False),
+    rpn_proposal=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_radius=1,
+        pos_weight=-1,
+        max_num_grid=192,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=1000,
+        nms_post=1000,
+        max_num=1000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.03, nms=dict(type='nms', iou_thr=0.3), max_per_img=100))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=3665,
+    warmup_ratio=1.0 / 80,
+    step=[17, 23])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 25
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/grid_rcnn_gn_head_r50_fpn_2x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/grid_rcnn/grid_rcnn_gn_head_x101_32x4d_fpn_2x.py

@@ -0,0 +1,180 @@
+# model settings
+model = dict(
+    type='GridRCNN',
+    pretrained='open-mmlab://resnext101_32x4d',
+    backbone=dict(
+        type='ResNeXt',
+        depth=101,
+        groups=32,
+        base_width=4,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch'),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        loss_cls=dict(
+            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+        loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        with_reg=False,
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False),
+    grid_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    grid_head=dict(
+        type='GridHead',
+        grid_points=9,
+        num_convs=8,
+        in_channels=256,
+        point_feat_channels=64,
+        norm_cfg=dict(type='GN', num_groups=36),
+        loss_grid=dict(
+            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=15)))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        debug=False),
+    rpn_proposal=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_radius=1,
+        pos_weight=-1,
+        max_num_grid=192,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=1000,
+        nms_post=1000,
+        max_num=1000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.03, nms=dict(type='nms', iou_thr=0.3), max_per_img=100))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=3665,
+    warmup_ratio=1.0 / 80,
+    step=[17, 23])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 25
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/grid_rcnn_gn_head_x101_32x4d_fpn_2x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]