Tianpeng LI

The optflow results of Pretrained weights is all zeros

10

预训练的SegFlow.caffemodel模型中的score_res3_con_flow层权值全为０导致光流输出结果全是０，希望能提供正确的权值下载 In pretrained weights SegFlow.caffemodel, weights of layer("score_res3_con_flow") is all zeros, please upload a correct pretrained weights, thank you!

main.py line 987 should be origin code will multiply unfinished_sequences with a array which elements not only the 0 or 1 for example eos_token_id = [1,2] next_tokens = torch.LongTensor([1,2,3,4,5]) sum(next_tokens...