Problems about training results of multimodal_context

ai4r / Gesture-Generation-from-Trimodal-Context

Speech Gesture Generation from the Trimodal Context of Text, Audio, and Speaker Identity (SIGGRAPH Asia 2020)

Other

245 stars 35 forks source link

Problems about training results of multimodal_context #53

Closed Jacoo-ai closed 8 months ago

Jacoo-ai commented 9 months ago

Hi, I met some problems about the training results of multimodal_context. I ran "python scripts/train.py --config=config/multimodal_context.yml" directly without modifying your original code. However, the results showed that the FGD is 52.480 at epoch 96, which is much higher than your results discussed in the paper (3.729). Is there something I missed? Should I modify the parameters in the config file to reproduce your result?

youngwoo-yoon commented 9 months ago

Hello, No, the project should reproduce similar FGD without modifying config or code. I have confirmed this by several runs when I release the code. But running environment might affect the results.