康行天下

I can get equal GPU/CPU result by your example. ```python import numpy as np from nms import nms, gpu_nms dets = np.array([[1.54879288e+02, 5.63751030e+01, 2.00795334e+02, 1.76900604e+02, 9.13123965e-01], [1.54429581e+02, 5.55528603e+01, 2.00706253e+02, 1.78303726e+02,...

I once used the file `test_augment.cpp` just for temporary testing. You can delete this file or rename it. Sorry for that. By the way, I have never run `make test`,...

I can compile out the `demo_yolov3.bin` in my machine with `make`, maybe you can compile again or modify the `Makefile`.

The code about yolo layer is copied from `pjreddie/darknet`, and you can find the xx_gpu functions at that repo. I didn't include those gpu code cause I need time to...

The input data format this repo support is as follows: ``` # image-id image-name number of boxes label x1 y1 x2 y2 difficulty ... ``` So if you like to...

Try to use `examples/FRCNN/generate_anchors.py` to generate anchors info.

You can adjust the settings in voc `cfg` file to adapt to different dataset.

You can set params in function `generate_anchors(base_size=16, ratios=[0.5, 1, 2], scales=2**np.arange(3, 6))`.

Yes, the GPU implementation in frcnn_proposal_layer.cu is not that efficient.

Yes, you are right. The Soft-NMS operation is similar with this.