facebookresearch / maskrcnn-benchmark

Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.
MIT License
9.3k stars 2.49k forks source link

Faster RCNN using ROI Pooling in maskrcnn-benchmark? #433

Closed mr2901 closed 5 years ago

mr2901 commented 5 years ago

❓ Questions and Help

I want to use Faster RCNN for training with ROI Pooling. I use this command: python tools/train_net.py --config-file "configs/e2e_faster_rcnn_R_50_FPN_1x.yaml" SOLVER.IMS_PER_BATCH 2 SOLVER.BASE_LR 0.0025 SOLVER.MAX_ITER 700000 SOLVER.STEPS "(480000, 640000)" TEST.IMS_PER_BATCH 1

I think the default setting is ROI Align instead of ROI Pooling. How can i use ROI Pooling?

Thanks.

fmassa commented 5 years ago

Hi,

I hard-coded that we would use RoiAlign in https://github.com/facebookresearch/maskrcnn-benchmark/blob/13b4f82efd953276b24ce01f0fd1cd08f94fbaf8/maskrcnn_benchmark/modeling/poolers.py#L66-L68

If you want to use Roi Pool, just change that line with something like

ROIPool(
    output_size, spatial_scale=scale
)

(and import ROIPool from layers) and it should be fine.

Additionally, you could add an extra config argument to select between ROIPool or ROIAlign.