cltdevelop

模型收敛问题

5

你好，想问一下，我在中文数据集上训练后，loss从150左右降到35， 但模型对于不同图片的预测结果始终是相同的37个汉字，比如“并并并.......”，能请教一下可能是什么问题吗？或者是什么问题导致模型不收敛呢？

endpoints.predicted_text outputs []

1

Hello! Thank you for your sharing. When I run your code, I find that the endpoint.predicted_text outputs nothing, is it right?

bucket_specs

2

Hello! Thank for your opening source code. It's a wonderful code. I want to ask what the meaning of bucket_specs in DataGen class: self.bucket_specs = [(int(64 / 4), 9 +...

support multi-gpu and batch image inference ?

1

Thank you for sharing your code! This is a wonderful work! I want to ask, does this code infer_simply.py support batch image inference and multi-gpu? Thank you in advance.

out,state = tflib.ops.FreeRunIm2LatexAttention('AttLSTM',emb_seqs,ctx,EMB_DIM,ENC_DIM,DEC_DIM,D,H,W)

2

Hello! Thank you for sharing your code! I have something wrong when I run this code. Is there something wrong in the attention.py. The order of parameters in out,state =...

Hello, when I run this code, there is a problem: /home/yzl/anaconda2/lib/python2.7/site-packages/h5py/__init__.py:34: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be...