森侃

Results 1 issues of 森侃

请问作者在做这个项目的时候有没有进行特征可视化,我最近使用grad cam对其做特征可视化,但是效果不太好,在路径聚合网络中的特征可视化很差,基本无法理解,我只找到两个效果比较好的层,主要是sppf前一层卷积backbone.image_model.stage4.2.conv2.conv,和bbox_head.head_module.cls_preds.2.1.conv。想问一下这种用yolo-world-x-1280模型跑出来的特征图效果就是这样的吗,期待作者的回复 ![Image](https://github.com/user-attachments/assets/81ed76f7-baac-47ab-9273-43232b80f14b) ![Image](https://github.com/user-attachments/assets/41814f49-dbee-44cc-8ff6-ea50fa09bc65) ![Image](https://github.com/user-attachments/assets/a5de4c90-a5b9-4cf5-8850-d6d812d1227a) ![Image](https://github.com/user-attachments/assets/ecc495bb-52c4-444a-b1a7-abb57f9326f8) ![Image](https://github.com/user-attachments/assets/b2c04cd2-5237-4e65-a152-7afa2bb6a295) ![Image](https://github.com/user-attachments/assets/0fb8f558-1b24-4fbd-a300-46af7ace25fc) ![Image](https://github.com/user-attachments/assets/8b45cbca-386d-4471-8e4e-e8b6bcd5ed77) ![Image](https://github.com/user-attachments/assets/23210e92-20df-4fd3-8450-891fa4c91c1f) # Copyright (c) Tencent Inc. All rights reserved. import os import cv2 import argparse import os.path as osp import torch...