[POC] Mask RCNN implementation to support PyTorch / XLA

[miladm]

Run Mask RCNN on PyTorch/XLA.

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit c8f7a12 (more details on the Dr. CI page):

---

• 17/17 failures introduced in this PR

---

🕵️ 2 new failures recognized by patterns

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

cmake_linux_cpu (1/2)

Step: "packaging/build_cmake.sh" (full log | diagnosis details | 🔁 rerun)

https://wiki.centos.org/yum-errors

(2/6): epel/x86_64/updatei 0% [                 ]  0.0 B/s |    0 B   --:-- ETA 
http://mirror.es.its.nyu.edu/epel/7/x86_64/repodata/7054bc90fda978065ad38663c2e7d148dd573288bb5a1622e06734322e84b8ac-updateinfo.xml.bz2: [Errno 12] Timeout on http://mirror.es.its.nyu.edu/epel/7/x86_64/repodata/7054bc90fda978065ad38663c2e7d148dd573288bb5a1622e06734322e84b8ac-updateinfo.xml.bz2: (28, 'Connection timed out after 30000 milliseconds')
Trying other mirror.

(2/6): updates/7/x86_64/primary_db                         |  13 MB   00:00     

(3/6): ius/x86_64/primary                                  | 100 kB   00:00     

(4/6): okay/7/x86_64/primary_db                            | 3.8 MB   00:00     

epel/x86_64/updateinfo         FAILED                                          

(5/6): epel/x86_64/primary 0% [                 ]  0.0 B/s |    0 B   --:-- ETA 
http://mirrors.vcea.wsu.edu/epel/7/x86_64/repodata/7054bc90fda978065ad38663c2e7d148dd573288bb5a1622e06734322e84b8ac-updateinfo.xml.bz2: [Errno 14] HTTP Error 404 - Not Found
Trying other mirror.
To address this issue please refer to the below wiki article 

https://wiki.centos.org/yum-errors

If above article doesn't help to resolve this issue please use https://bugs.centos.org/.


epel/x86_64/updateinfo         FAILED                                          

(5/6): epel/x86_64/primary 0% [                 ]  0.0 B/s |    0 B   --:-- ETA 
https://d2lzkl7pfhq30w.cloudfront.net/pub/epel/7/x86_64/repodata/7054bc90fda978065ad38663c2e7d148dd573288bb5a1622e06734322e84b8ac-updateinfo.xml.bz2: [Errno 14] HTTPS Error 404 - Not Found
Trying other mirror.

epel/x86_64/updateinfo         FAILED                                          

(5/6): epel/x86_64/primary 0% [                 ]  0.0 B/s |    0 B   --:-- ETA 
https://dl.fedoraproject.org/pub/epel/7/x86_64/repodata/7054bc90fda978065ad38663c2e7d148dd573288bb5a1622e06734322e84b8ac-updateinfo.xml.bz2: [Errno 14] HTTPS Error 404 - Not Found
Trying other mirror.

epel/x86_64/updateinfo         FAILED                                          

type_check_python (2/2)

Step: "Check Python types statically" (full log | diagnosis details | 🔁 rerun)

Found 7 errors in 2 files (checked 226 source files)

            import torch_xla.debug.metrics as met
    ^
torchvision/models/detection/generalized_rcnn.py:122: error: "Tensor" not
callable  [operator]
            detections = self.transform.postprocess(detections_cpu, images...
                         ^
torchvision/models/detection/rpn.py:267: error: Argument 5 to "batched_nms" has
incompatible type "Callable[[], int]"; expected "int"  [arg-type]
    ....batched_nms(boxes, scores, lvl, self.nms_thresh, self.post_nms_top_n)
                                                         ^
Found 7 errors in 2 files (checked 226 source files)


Exited with code exit status 1

---

15 failures not recognized by patterns:

Job	Step	Action
unittest_linux_cpu_py3.8	Run tests	🔁 rerun
unittest_windows_cpu_py3.7	Run tests	🔁 rerun
binary_linux_conda_py3.9_cu102	packaging/build_conda.sh	🔁 rerun
unittest_linux_cpu_py3.7	Run tests	🔁 rerun
unittest_onnx	Run tests	🔁 rerun
unittest_linux_cpu_py3.9	Run tests	🔁 rerun
unittest_torchhub	Run tests	🔁 rerun
lint_c	Lint C code	🔁 rerun
unittest_windows_cpu_py3.9	Run tests	🔁 rerun
binary_linux_conda_py3.7_cpu	packaging/build_conda.sh	🔁 rerun
lint_python_and_config	Lint Python code and config files	🔁 rerun
binary_linux_conda_py3.9_cpu	packaging/build_conda.sh	🔁 rerun
unittest_windows_cpu_py3.8	Run tests	🔁 rerun
binary_linux_conda_py3.8_cpu	packaging/build_conda.sh	🔁 rerun
unittest_prototype	Run tests	🔁 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.

[miladm]

The current implementation successfully runs on PyTorch/XLA.

Some performance optimizations to hand dynamic shape remain to be supported after this feature is available on PyTorch/XLA.