fighterzzzh

When I perform image matching with two images, obtaining the ROI after matching and obtaining the ROI before matching, the former yielded better results. Why is this the case?

The Difference Between "forward_symmetric" and "forward"

4

Hello, great job! I have a question to ask. What is the difference between "forward_symmetric" and "forward"? Looking forward to your reply.

Hello, I am using SuperPoint+LightGlue for matching, but the performance on port containers is not very good. I heard that COLMAP can be used to obtain the necessary training data,...

train with my own video

4

Hello, I would like to ask if I need to train with my own video? My video contains my own scenes, but the camera might not be able to recognize...

lightglue实际部署很差，是否是哪里有误

3

大佬你好，我把lightglue部署后正常，但GIM的模型却很差，只有一个特征点，我估计是哪有问题，我在转onnx有如下处理会有问题吗？ ``` import torch #-假设检查点文件包含一个名为 state_dict 的模型状态字典 checkpoint = torch.load(R"D:\pycharm\gim\weights\gim_lightglue_100h.ckpt", map_location="cpu") state_dict = checkpoint['state_dict'] #-将 state_dict 保存为 .pth 文件 torch.save(state_dict, 'gim_lightglue_100h.pth') ``` ``` #---------------------调用时---------------------------- state_dict = torch.load( R"D:\pycharm\LightGlue-ONNX\gim_lightglue_100h.pth" ) ```...

很出色的作品，请问多图处理的时候是将图像同时输入进去了 ，这样的优势是什么？

1

举个例子，比较两张图像的异同，我可以这样做： 1. 图像 1 送入vllm获取描述1 2. 图像2 送入vllm获取描述2 3. 描述1和2送入llm获取异同 也可以采用多图输入 进行比较，这两种方法有什么不同吗，他们的实现有什么不同吗？第一种方法就是多轮对话吗？

[Bug] lora微调internvl2.58B并合并后，无法推理，报错

1

### Checklist - [x] 1. I have searched related issues but cannot get the expected help. - [ ] 2. The bug has not been fixed in the latest version....

[Feature] 视觉大模型中视觉部分通过tensorrt

1

### Motivation 你好！ 请问视觉大模型，例如internvl，考虑过将视觉部分通过tensorrt进行推理吗，我预计这会产生更快的加速? ### Related resources _No response_ ### Additional context _No response_

Hello, may I ask if it's not possible to train if I only have a sparse depth map?