search

hanzhanggit / StackGAN-Pytorch

MIT License
491 stars 146 forks source link

inception score on coco pretrained model is 8.50526228697 #6

Closed kracwarlock closed 5 years ago

kracwarlock commented 6 years ago

@hanzhanggit The readme says

Our current implementation has a higher inception score(10.62±0.19) than reported in the StackGAN paper

With the provided pre-trained model, I get 8.50526228697. Is the 10+ score for StackGAN v2?

hanzhanggit commented 6 years ago

@kracwarlock No. The inception score (10.62±0.19) is calculated from the Pytorch model of stackgan-v1

kracwarlock commented 6 years ago

I was unable to get that from the available pre-trained model in the repo. Can you share the evaluation code for the pytorch model?

How many images did you calculate this over and what were the splits?

AnwarUllahKhan commented 6 years ago

@kracwarlock @hanzhanggit @taoxugit Please tell me the path details.

I put my coco train images into data/coco folder so the complete path is "H://StackGAN-Pytorch/data/coco" and put the preprocessed char-CNN-RNN text embeddings training coco, and saved it to data/coco/train and test folders.... complete path is H://StackGAN-Pytorch/data/coco/train" and H://StackGAN-Pytorch/data/coco/test" As well as put the preprocessed char-CNN-RNN text embeddings evaluating coco, and save it to data folder complete path is H://StackGAN-Pytorch/data/coco/test"

now tell me these paths are ok or not? and where I put the coco val images? Are any other things are needed?

Strongly waiting for your's response.... thank you very much