ActivityNet

2016 ActivityNet action recognition challenge. CNN + LSTM approach. Multi-threaded loading.

Notes:

Nice example of training w/ and w/o multi-threaded loading
Did not finish before challenge deadline (no test results)
Bug in LSTM forced use of regular RNN (investigate error & fix if re-using)
Need to find a faster way to load videos with lua (currently slower than training)

Name	Name	Last commit message	Last commit date
Latest commit History 4 Commits 4 Commits
README.md	README.md
VideoHFlip.lua	VideoHFlip.lua
VideoNormalize.lua	VideoNormalize.lua
VideoOptFlow.lua	VideoOptFlow.lua
VideoSpatialCrop.lua	VideoSpatialCrop.lua
dataset.lua	dataset.lua
main.lua	main.lua
model.lua	model.lua
printer.lua	printer.lua
train.lua	train.lua
trainThreads.lua	trainThreads.lua
validation.lua	validation.lua

Provide feedback