hbai98 / scm Goto Github PK

Hi! I really like the your work and I tried that on my RTX 3060 12G, but it kept going NaN so I added lr scheduler and refactor the whole codebase. Here is the link https://github.com/author31/scm_nbdev. Once again, thanks for your work, keep doing.

Hello! I read your article carefully and was very interested in it! I have some questions as follows:

(1) Does the semantic similarity matrix E calculate the semantic similarity between all patchs?
(2) After I print E, I find a negative value. What does a negative value in E mean? (For example,the negative value -0.0383 in the first row)
tensor([[[ 1.0000, 0.3413, 0.3903, ..., 0.1250, -0.0383, 0.1996],
[ 0.3413, 1.0000, 0.4638, ..., 0.0055, 0.0692, 0.2095],
[ 0.3903, 0.4638, 1.0000, ..., 0.0800, -0.1332, 0.2198],
...,

(3) Does SCM diffuse only according to the semantic and spatial relations of the four points of its first-order neighbors?

Hope to get your reply! Thank you very much!

PatchCAM refers to the CAM generated by patch token. I feel strange about the activation of patchCAM on the CUB dataset, it seems to have higher activation for the whole image (almost the whole CAM is red). But in particular, the activation of the foreground object is relatively weaker than that of the background. patchCAM does not seem to provide class-specific semantic information on CUB. But on ImageNet, patchCAM seems to be normal again, such as strong foreground object activation and weak background activation, which can provide category-specific semantic information. I wonder why such an interesting phenomenon occurs.

Hello, I am a second year software engineering master student and I am poor at math.

I have some questions about the equations. In section 3.2, "(Ll)−1 i,j describes the correlation of vli and vlj at the equilibrium status", I don't understand the relationship between "inverse of Laplacian matrix" and "equilibrium status". Besides, In the appendix, I don't know how to get Eq.(4) from Eq.(2).

I wonder which books or papers should I refer to for understanding this equations better? Could you provide us with the information?
Thanks.

Thanks for sharing your wonderful work!
Could you offer me these files?

Hello! First of all, your work is excellent! Congratulations on a good result! Second, I'm very interested in your work, and there's a problem running the code.

The results cls_acc_top-1、cls_acc_top-5、loc_acc_top-1、loc_acc_top-5 and GT_Known using the source code on CUB are different from those in the paper. Compared to the paper results, I used the same settings, but the results decreased to varying degrees about 3%~4%.

Hope to get your reply and answer, thank you!