SpatioTemporalEntropyModel
SpatioTemporalEntropyModel copied to clipboard
Spatio-Temporal Entropy Model (STEM) for end-to-end leaned video compression.
When training and testing the spatio-temporal entropy model, is the I-frame training model using a pre-trained model? Another question, args.i_model_path I did not find, is as follows 
大佬请问一下 一直没想明白为啥他的编码端会比解码端FLOPs低,感觉只有熵模型的话,编码端解码端计算量应该差不多吧 还有那个测编码端的FLOPs是,I帧网络的g_a部分 + 熵编码的compress部分,两部分计算量相加吗  感谢大佬分享,感激不尽orz