Yihao Chen

Hello sir. I have one problem, my training loss can't converge when it is lower then 3.00 . I trained with 3,800K data (crop from SynthText)， lr = 0.001, adam...

@gachiemchiep thanks a lot. I will try it as you suggest

@gachiemchiep I have sloved my problem， I found that warp-ctc cannot work with multi GPU, so I set ngpu=1 and it works profect [https://github.com/SeanNaren/warp-ctc/issues/78](url)

> 将中文数据的中文字符当成同一个字符即可 你好，我好奇你们的训练集是生成的还是真实的？数据量有多少？中英文识别的效果都很不错

Hi, thanks for attention, 'num_features' means the number of node/edge feature channel in this work. and you can see more about 'num_features' in BatchNorm, InstanceNorm: [https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm1d.html](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm1d.html) [https://pytorch.org/docs/stable/generated/torch.nn.InstanceNorm1d.html#torch.nn.InstanceNorm1d](https://pytorch.org/docs/stable/generated/torch.nn.InstanceNorm1d.html#torch.nn.InstanceNorm1d)

You can modify it in config file, such as ['configs/SBMs_node_clustering_GCN_CLUSTER_100k.json'](https://github.com/cyh1112/GraphNormalization/blob/master/configs/SBMs_node_clustering_GCN_CLUSTER_100k.json) line 27: net_params.hidden_dim

Hi, @Sharathmk99 Thanks for following our work. 1. How can i change to something like work2vec or tfidf? Yes, that's a good idea. please show me your results if you...

@enhenghengheng 能不能提供一下你的环境信息，包含 Liunx, Python, pytorch, mmcv, cuda, cudnn, gcc等

@enhenghengheng mmcv 换成0.2.14，然后重新安装future和tensorboard试试

@enhenghengheng 从这个错误信息来看，环境应该是没问题了，新的报错就是图片找不到，请好好检查一下路径