wenzhi fang

about RGB and BGR

1

"imgA = cv2.imread('last/'+img_name)" my doubt is that the input data's format is BGR but VGG is compatible to RGB

请教一下 我认为您这样做会出现一个像素点label为【1，1】或【0，0】的情况啊 还有您的代码似乎做的是对通道的分类 不是对像素点的分类 比如对224*224*3的图片，label应该是224*224*n_classes 吧 但是您的是224*224*3*n_classes

question

1

i am going to learn your technique to coding. for me, i feel your data preprocess is so difficult.

memory issue, prepare_model_for_kbit_training

4

I am using a quantized base model (NF4) and do LoRA fine-tuning. When I call prepare_model_for_kbit_training to wrap the model, the memory consumption is significantly higher than the bf16 counterpart,...