search

chenjoya / 2dtan

An optimized re-implementation for 2D-TAN: Learning 2D Temporal Localization Networks for Moment Localization with Natural Language (AAAI'2020).
125 stars 15 forks source link

Where is the improvement come from? #2

Closed frostinassiky closed 4 years ago

frostinassiky commented 4 years ago

Hi Joya,

Thanks again for sharing your brilliant work!

From your experience, where is the improvement come from? Why does the same architecture (2D-TAN) give very different performances?

chenjoya commented 4 years ago

Hi Frostinassiky,

Our clean implementation speeds up the training and inference. I think that the improvement on recall@k comes from the learning hyper-parameters (learning rate, milestone). We use larger learning rate and less training epoches.

Thanks for your question. Please feel free to contact me if you have other problems.

frostinassiky commented 4 years ago

Thanks for your response! Again it is very impressive to see the boost!