Xiaoyang
Xiaoyang
@nemonameless 你好，我也出现了这个问题，loss一直在下降，验证集的准确率也在不断上升，但是测试集就崩了，全是0.5左右…… 请问你找到问题所在了么？
我找到问题所在了：README文档给出的测试命令不对，预训练模型应该使用`--load-model-path`参数，而不是`--load-path`。
Hi, I encountered the same question, could you Explain how to solve it? My codes are like this: ``` total_lpips, total_lpips_convert = 0, 0 with torch.no_grad(): loss_fn_vgg = lpips.LPIPS(net='vgg').cuda() for...
You can refer to: https://github.com/oreillymedia/t-SNE-tutorial. I hope it helps.
Hi @lecanhduc02 , currently we do not have the code for colab training. Here is a colab example of inference that might be helpful：https://github.com/camenduru/DDColor-colab
Yes, it is possible to use the modelscope library for training. However, the reliability of this method (i.e., the alignment of the final results with our model results) is not...
Hi @liangbaikaizzzZZZ ! Since DDColor is not specifically trained for video colorization, so it is normal for sub-optimal results. However, you can retrain a model and use early-step checkpoints for...
Sorry, I didn't get it. data_list/get_meta_file.py is a sample script for generating the data list file needed for training. Before starting the training, you need to ensure that all data...
Hi, @weixin6666 , imagenet_val_5k is actually the first 5,000 images in imagenet validation set, you can edit the script to generate it.
Hi @matureshadow, according to the error message, it may be a problem with installing dlib on windows. Perhaps you can refer to this: https://stackoverflow.com/a/76630254