add FCOS

[xiaohu2015]

The pr is about #2707
We (@xiaohu2015 and @zhiqwang ) implemented torchvision-style FCOS by referencing to detectron2 and AdelaiDet repo.

if train with the retinanet training command:

torchrun --nproc_per_node=8 train.py\
    --dataset coco --model fcos_resnet50_fpn --epochs 26 \
    --lr-steps 16 22 --aspect-ratio-group-factor 3 --lr 0.01 --amp

you can get:

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.390
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.585
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.415
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.221
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.419
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.511
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.325
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.526
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.562
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.359
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.603
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.717

the AP result is better to the paper (38.6).

if train with training command (1x, 12 epoch, the dafult setting in the FCOS paper):

torchrun --nproc_per_node=8 train.py\
    --dataset coco --model fcos_resnet50_fpn --epochs 12 \
    --lr-steps 8 11 --aspect-ratio-group-factor 3 --lr 0.01 --amp

you can get:

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.385
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.583
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.411
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.225
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.424
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.490
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.325
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.533
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.572
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.384
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.619
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.709

the AP result is smilar to the paper (38.6).

cc @jdsgomes

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 6840a7d (more details on the Dr. CI page):

---

• 1/2 failures introduced in this PR
• 1/2 broken upstream at merge base fe65d37 since Jan 14

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

unittest_linux_gpu_py3.8 (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

test/test_utils.py::test_draw_keypoints_errors �[33mSKIPPED�[0m (We're in a C...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks[1-blue] �[33mSKIPPED�[0m (We...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks[1-#FF00FF] �[33mSKIPPED�[0m�[31m      [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks[1-colors3] �[33mSKIPPED�[0m�[31m      [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks[1-colors4] �[33mSKIPPED�[0m�[31m      [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks[1-colors5] �[33mSKIPPED�[0m�[31m      [ 99%]�[0m

test/test_utils.py::test_draw_segmentation_masks_errors �[33mSKIPPED�[0m (We'...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_keypoints_vanilla �[33mSKIPPED�[0m (We're in a ...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_keypoints_colored[red] �[33mSKIPPED�[0m (We're ...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_keypoints_colored[#FF00FF] �[33mSKIPPED�[0m (We...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_keypoints_colored[colors2] �[33mSKIPPED�[0m (We...)�[31m [ 99%]�[0m

test/test_utils.py::test_draw_keypoints_errors �[33mSKIPPED�[0m (We're in a C...)�[31m [ 99%]�[0m

test/test_video_gpu_decoder.py::TestVideoGPUDecoder::test_frame_reading �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_stress_test_read_video_from_file �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_read_single_stream_only �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_min_dimension �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_max_dimension �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_both_min_max_dimension �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_width �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_height �[33mSKIPPED�[0m�[31m [ 99%]�[0m

test/test_video_reader.py::TestVideoReader::test_read_video_from_file_rescale_width_and_height �[33mSKIPPED�[0m�[31m [ 99%]�[0m

---

🚧 1 ongoing upstream failure:

These were probably caused by upstream breakages that are not fixed yet.

• unittest_prototype since Jan 14 (adf8466)
  • 🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[jdsgomes]

Thank you for all your patience and commitment. Looks great to me now and I think its ready to merge, but I will just wait for @datumbox approval to make sure I didn't miss any changes suggested by him.

[datumbox]

LGTM! Thanks a lot @xiaohu2015 and @zhiqwang for your amazing work.

We are going to base our new model contribution process on this collaboration. @jdsgomes is currently leading this and we will ping you once this is drafted to get your input.

Looking forward working with you again, soon. Perhaps next target is EfficientNetV2? 😃

[github-actions]

Hey @jdsgomes!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py

[jdsgomes]

I have opened a ticket to add FCOS model in Pytorch Hub: pytorch/hub#265