Open graphl opened 1 year ago
Use .load_state_dict
before the training (e.g, in this place). Do not use .eval
because if is for training. If your dataset is small, it is more like finetune and you need only train for a few steps.
Thank you for the reply,I've tried it。The method you said is very useful. But I ran into other problems。 I trained with epoch = 11 and added some of my own fonts. But test with the parameters of their own training, but the effect is not very good.
Here are some of my test results。
I found that the weight of the font seemed to affect the result.
I found that the thickness of the font seems to affect the result, the thicker the font, the better the effect, and the thinner the font, the worse the effect。What do I have to do to make the thin font better。
Hi, thank you for making this! I loaded the parameters of the pre-trained model and trained it, But now the results more like the fonts I added (main.py)Append code:
result![merge_128](https://user-images.githubusercontent.com/33078222/216914470-b5d302b9-b8dd-44e1-937d-7f945089fd95.png)
VAL/Loss![image](https://user-images.githubusercontent.com/33078222/216923051-6d0dbd4d-f781-4b97-8cb2-e60a10e4ec0a.png)
Is this an overfitting? or This is because there are no parameters to load the Adam optimizer ? If I want to continue training my method on the basis of pre-trained, right?