vae_captioning
vae_captioning copied to clipboard
can you give more details about the ag_cvae or other models' parameters in training and testing?
Or maybe just use the default? I got cider=0.95 when I set model=AG_CVAE, no finetune; but when I use finetune, the cider turn down to 0.85. When I change the feature from VGG_16(dim=4096) to Resnet101(dim = 2048), the cider score down to 0.85