Huanmin Dou

Results 19 comments of Huanmin Dou

感谢您的提醒,我要去读原版了🙏

你好,我这边有一个tsne的,也是做哈希用的,可以参考一下,对于多标签数据集,我一般只选了单标签数据集进行tsne可视化 https://github.com/Mahiro2211/T-SNE_visualization

![2023-11-06 20-04-03 的屏幕截图](https://github.com/ggjy/CMT.pytorch/assets/130811701/2e51d33e-7c65-455f-b67d-5ecc514c5243)

是可以的,只需要按照他的函数的输入形状设计你的特征就可以了。我之前用在过跨模态检索上一次,但是效果不是很好

> > 我也想要一个 > > 拿到了吗 我也想要 他的这个可视化应该是在ViT的cls做的,我觉得应该和这个差不多,[博客链接](https://blog.csdn.net/z240626191s/article/details/132504292?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522171289468016800185871958%2522%252C%2522scm%2522%253A%252220140713.130102334.pc%255Fall.%2522%257D&request_id=171289468016800185871958&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~first_rank_ecpm_v1~rank_v31_ecpm-1-132504292-null-null.142^v100^pc_search_result_base1&utm_term=ViT%20cls%E5%8F%AF%E8%A7%86%E5%8C%96&spm=1018.2226.3001.4187) 但是在224x224的图片上有点糊,最后还是用了CAM方法来可视化了

Thank you for sharing the code, It is important not to ignore the output that hasn't been flattened

I found that if I deal with the coco dataset with the same method of dealing with nuswide , It works very well I changed the config ```python 'dataset_kwargs': {...