XuShoweR

Results 3 issues of XuShoweR

I want use your triplet_loss function to train a classification model with imagenet, and i got a mistake at triplet_loss.py line 70 dist_ap, relative_p_inds = torch.max((dist_mat * is_pos.float()).contiguous().view(N, -1), 1,keepdim=True)...

So all the pictures with fog in the training set are synthetic? If I have paired images , how can I use them

我现在在做基于qwenvl的grpo,发现在做Grounding任务的时候如果输入相对小点的分辨率回答正常以thinking开头 并且loss正常,而当我使用稍大一点的图片那回答必然会重复一遍问题 而且loss还很大。训小尺度就没这个问题。像素也在max_pixels=12845056 min_pixels=3136 之间,大概是1280*1080,请问这是什么原因,是我有哪里没设置好吗 正常训练指标 ![Image](https://github.com/user-attachments/assets/ef19fdcd-0845-40a2-9581-a70d9e3d992d) 大尺度,不正常指标 ![Image](https://github.com/user-attachments/assets/bdc3329c-e853-458b-a22a-486bc3864aab) 两个实验只有分辨率不一样。当尺度大了之后iou基本在0.4左右,小尺度反而可以到0.8左右