search

yytzsy / SCDM

Code for the paper: Semantic Conditioned Dynamic Modulation for Temporal Sentence Grounding in Videos
68 stars 14 forks source link

Questions about the pretrained model for ActivityNet #4

Closed syuqings closed 4 years ago

syuqings commented 4 years ago

Hi, Yitian, I loaded the pretrained checkpoint of ActivityNet dataset but failed. The vocab size in the checkpoint is about 14000, while the word2ix.npy generated by train() is about 21000. Do you know what's the problem of it or how to get the corresponding vocabulary? Thanks.

yytzsy commented 4 years ago

Maybe you should first run the train() process and generate the corresponding vocabulary for the ActitvtiyNet Captions dataset, and then load the pretrained model for test.

syuqings commented 4 years ago

Yes, I have generated the vocabulary by running the train() process, but the vocabulary size is about 21000 which is not match with the checkpoint file.

yytzsy commented 4 years ago

I will check the checkpoint, and you can train the model by your own first and reproduce the results in the paper.

syuqings commented 4 years ago

OK, Thanks