Hongwei Niu

[Question] Recognition Error

2

### Question Hello, I found that it, like CLIP, focuses on a very coarse granularity. For example, an image cropped from the GT box of the COCO dataset is recognized...

Can you share some implementation details about the result about 'K-Means Clustering of Frozen Features'?

Hello, I've been working on a project involving object detection, and I encountered a specific issue I'd like to discuss. My approach involves using RoIalign to extract regional features from...

关于代码中y的疑问

4

你好，我想知道torch_vertex.py 146行，y代表什么，有什么作用呢？ ``` if self.r > 1: y = F.avg_pool2d(x, kernel_size=self.r, stride=self.r) # [B, out_dim, H/r, W/r] y = y.reshape(B, C, -1, 1).contiguous() # [B, out_dim, H/r*W/r, 1] ```

What do Pair, L, T stand for in the code?

3

Hi, I'm a beginner and would like to ask a question. What do Pair, L, T stand for in the code? What do they mean? ``` # Pair x L...

关于MultiModalDataset

1

作者，您好。请问对于同义词你是怎么处理的呢？像ODISE、FC-CLIP的做法是在预测的时候进行max ensemble，但是这样的操作特别耗时，会影响FPS。 例如，num_templates存储的是每个类别的同义词个数，它会将每个类别的所有同义词模板进行max操作，这样得到的final_pred_logits维度为[B, N, num_classes] ``` cur_idx = 0 for num_t in num_templates: final_pred_logits.append(pred_logits[:, :, cur_idx: cur_idx + num_t].max(-1).values) cur_idx += num_t ```

CLIP Recognition Error

2

CLIP will recognize this image as a hot dog with a very high probability close to 1, but the actual label should be a person. Is there a solution? 

AttributeError: module 'signal' has no attribute 'SIGUSR1'

1

hi, I have a problem: ```configs/faceshq_vqgan.yaml -t True --gpus 0, Running on GPUs 0, Working with z of shape (1, 256, 16, 16) = 65536 dimensions. loaded pretrained LPIPS loss...

Questions about input pred label file

3

Why is only one category displayed in the input pred label file?

Some questions about QSAttn.

8

(1) When computing category codes, src is support feature, why do src, category_code and tsp need to do QSAttn? It's not mentioned in the paper. (2) Why do QSAttn only...