search

msracver / Flow-Guided-Feature-Aggregation

Flow-Guided Feature Aggregation for Video Object Detection
MIT License
723 stars 191 forks source link

inference code is use i-2K frames to predict i but （-K to K） frames is used in paper，why difference？ #19

Closed dlongry closed 6 years ago

dlongry commented 6 years ago

OK，I misunderstanding the code，thanks for your work

Feywell commented 6 years ago

@dlongry How to understand the way of using frames in code. I have the puzzle as you. Can you tell us details?

dlongry commented 6 years ago

You should see the "get_aggregation_symbol" function in the "\fgfa_rfcn\symbols\resnet_v1_101_flownet_rfcn.py": such as line 1057 ， use cur_data = mx.symbol.slice_axis(data_cache, axis=0, begin=cfg.TEST.KEY_FRAME_INTERVAL, end=cfg.TEST.KEY_FRAME_INTERVAL+1) to get middle frame to detect。 @Feywell

Feywell commented 6 years ago

@dlongry Thank you. So, it means it use frames from -k to k as paper.