xxxxxxxiao

Results 9 comments of xxxxxxxiao

Hi! have u fixed the problem? i got the same issue. My loss was able to converge to 1.xx but when i trained it again i got 4.9xx. I believe...

setting a smaller learning rate may work

> 这是densenet测试了十几万张图片之后报的错 > 2019-01-23 19:51:42.943120: F tensorflow/stream_executor/cuda/cuda_dnn.cc:542] Check failed: cudnnSetTensorNdDescriptor(handle_.get(), elem_type, nd, dims.data(), strides.data()) == CUDNN_STATUS_SUCCESS (3 vs. 0)batch_descriptor: {count: 1 feature_map_count: 128 spatial: 4 0 value_min: 0.000000 value_max: 0.000000...

CRNN。 这样做的话会不会影响梯度下降呢?

貌似找到原因了。 val_gen = DataGenerator(val_list, img_size, down_sample_factor, batch_size, max_label_length) crnn_blstm_ctc.py中这一行的down_sample_factor, batch_size这两个参数位置颠倒了。。

能否问下你最后loss大概是多少?我改batchsize=64训练后测试集loss大概在0.77,但是demo输出的结果都是‘00’或者‘40’这种。不知是哪里操作不对?

我把你训练的模型加载后再训练,一开始出来的loss很高,请问有没有可能是loss的计算方式有问题?

验证集降到16,应该是跟你训练的时候一样啊。。不知道为什么val loss还是那么高。。。