lllllllllllllllllllllllllllllllllllllllllllllll

您好，因为我目前只有检测灰度图的需求，训练时所用数据集也全是灰度图，使用yolov7-tiny版本。请问有什么办法可以针对这种情况减少通道数，使得减少模型的参数量以达到提高训练和推理速度效果呢？如果有的话，可以讲解一下具体如何修改代码或配置文件吗？感谢！ Hello, because I only need to detect grayscale images at present, the datasets used in training are all grayscale images, using the yolov7-tiny version. Is there any way to...

Comment

1

What a wonderful theme! Any idea of making a series of 'haru-like' theme? (eg. blue, grey or some other colors instead of purple) Looking forward to see that 👀👍

I'm finetuning BLIP2 on coco, using the script `train_retrieval_coco.sh`, but only checkpoint of epoch 0 was saved. How to fix that? I want to save checkpoint after each epochs.

Hi authors, thanks for the great work! I want to finetune a BLIP2 model on my custom dataset to do text-image retrieval tasks. I finetuned model `Salesforce/blip2-opt-2.7b` on HuggingFace (loaded...