mzz

why get same output with different input?

4

@Lsdefine Thanks for your sharing, I use transformer to do seq2seq task. Like, input a article and predict the abstract. When I finish training, I get almost same output with...

ctpn/prepare_training_data/split_label.py 是否有问题？

2

你好，首先感谢分享代码！ chinese_ocr/ctpn/prepare_training_data/split_label.py 第49-70行这种策略是否有问题？ 比如对于一个平行四边形的label框，按照你的方法出来的xmin并不是xmin，并且你的方法分割出来的小框高度都是固定的，对于斜的平行四边形gt貌似不太友好？ 是不是应该参照text-detection-ctpn/utils/prepare/utils.py 中的shrink_poly这个方法来呢？ 请指教~

rcnn_box loss equals to 0 when training

5

@jwyang Hello, I trained fpn with resnext backbone on my own datasets and I get zero rcnn_box loss during the training, I wonder if there is something wrong ? meanwhile,...

Hello, Thx for your sharing. Do you have pretrained model?

I want to train model with lora, trainable params only include llm layers.*, I do not want to train "embed_tokens" and other layers 1）So I delete “embed_tokens” in modules_to_save list,...