yafeng

icon indicating a website link https://github.com/alibaba/MNN icon indicating an email [email protected]

icon company Alibaba icon location Hangzhou China icon location deep learning, model compression, AI infra

Results 3 issues of yafeng

Deep-Learning-Papers-Reading-Roadmap

Deep-Learning-Papers-Reading-Roadmap

RoiAlign with max pool, implementation differs from mmcv, need clarification

onnx: https://github.com/onnx/onnx/blob/main/onnx/reference/ops/op_roi_align.py#L235 mmcv: https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/csrc/pytorch/cpu/roi_align.cpp#L188 for the case: RoiAlign with max pool type onnx performs max pool with weighted four points in the source input, no sum operation, so no interpolated...

bug
topic: spec clarification
contributions welcome

what if set all layers' massive activations to their mean value?

1 comment

interesting work! i have a question as it in the title, do you conducte an experiment like that? what's the result? thanks.