View Code? Open in Web Editor
NEW
SOLO: Segmenting Objects by Locations
Python 83.87%
C++ 5.22%
Cuda 10.29%
Shell 0.16%
Cython 0.44%
solo's Introduction
๐ Hi, Iโm Yuqing Wang, a PhD student in the University of Hong Kong.
๐ Iโm interested in computer vision and generative models.
๐ซ How to reach me : [email protected] .
solo's People
Contributors
solo's Issues
Could you please tell me how to compile .cu and .cpp file in mmdet/ops/xxx/src
I am not familiar with this
thanks
I don't know much about mmdetection. where can I set batch_size in r50_p6.py?
There seems to be a bug in the below line.
iou = float (overlap .sum ())+ 1 / (float (union .sum ())+ 1 )
Might it be the following line?
iou = float (overlap .sum () + 1 ) / float (union .sum () + 1 )
And it might be faster to apply mask-nms on gpu.
Hi~ thanks for ur work
I notice that in ur implement, all grids touching gt center regions are sampled as positive
left_ = np .int (left [i ]// w [j ])
which is different from original paper 'grid (i, j) is considered as a positive sample if it falls into the center region of any ground truth mask'.
By my understanding, it's a positive sample as the grid center falls in center region.
Did u compare the net forward speed compares to maskrcnn or other one-stage instance segmentation methods?