detectron dependency in tools

[botcs]

🐛 Bug

As suggested in the data/README.md tools/convert_cityscapes_to_coco.py requires detectron to be installed.

To Reproduce

Steps to reproduce the behavior:

1. Install maskrcnn_benchmark following the INSTALL.md
2. Install additional dependencies for the script: pip install h5py scipy opencv-python
3. run python tools/cityscapes/convert_cityscapes_to_coco.py

Error:

ModuleNotFoundError: No module named 'detectron'

Environment

PyTorch version: 1.0.0.dev20190207
Is debug build: No
CUDA used to build PyTorch: 10.0.130

OS: Ubuntu 16.04.5 LTS
GCC version: (Ubuntu 5.4.0-6ubuntu1~16.04.10) 5.4.0 20160609
CMake version: version 3.5.1

Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: Could not collect
GPU models and configuration:
GPU 0: GeForce RTX 2080
GPU 1: GeForce RTX 2080

Nvidia driver version: 410.72
cuDNN version: Could not collect

Versions of relevant libraries:
[pip] numpy==1.16.1
[pip] torch==1.0.0.dev20190207
[pip] torchvision==0.2.1
[conda] blas 1.0 mkl
[conda] mkl 2019.1 144
[conda] mkl_fft 1.0.10 py36ha843d7b_0
[conda] mkl_random 1.0.2 py36hd81dba3_0
[conda] pytorch-nightly 1.0.0.dev20190207 py3.6_cuda10.0.130_cudnn7.4.2_0 pytorch

[fmassa]

Oh, you are totally right, thanks for spotting this!

The dependency on Detectron is really not necessary, and was probably an oversight from #232

There are only a couple of places where detectron is needed, and only two functions are used, see

maskrcnn-benchmark/tools/cityscapes/convert_cityscapes_to_coco.py

Lines 196 to 198 in e60f4ec

	ann['bbox'] = bboxs_util.xyxy_to_xywh(
	segms_util.polys_to_boxes(
	[ann['segmentation']])).tolist()[0]

We already have something similar to bboxs_util.xyxy_to_xywh in BoxList via the convert method. The second function can be taken from Detectron to avoid this dependency.

Contributions are more than welcome! :-)

[fmassa]

Actually, I'd like to keep this open until the dependency on detectron is removed.

[botcs]

sorry, I have closed it accidentally.
I have a new PR on the detectron free implementation :)