Jianyuan Wang

Hey I just uploaded a new demo_colmap.py file. Hope it can solve the problem

Hi @xyyeah , Can you check the log showing the device and dtype? The script has the lines below: print(f"Using device: {device}") print(f"Using dtype: {dtype}") Normally it should show using...

Nx518x518x3 is designed because this script requires a square image

Hi @zzh-tech, we pad and resize the input images to square for the code in BA. It should be possible to revise the code to support non-square images but may...

Hey those noisy 3D points look like the predictions of the black background points. In our training data, we have images with pure black/white backgrounds and we do not apply...

Hi, we use the flash attention embeded in pytorch F.scaled_dot_product_attention. If you have installed flash attention by yourself, you need to replace the line below to your one https://github.com/facebookresearch/vggt/blob/c4b5da2d8592a33d52fb6c93af333ddf35b5bcb9/vggt/layers/attention.py#L61

Hi, it is not normal. I think it means flash attention cannot be applied. Basically it is saying the inbuilt function [scaled_dot_product_attention](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) is not working correctly. After a simple search,...

Hi @dsvilarkovic , thanks for your interest! In VGGSfM, we use Aliked/Sift/Superpoint to extract keypoints, and run tracking over those keypoints to get the tracks over other frames. In VGGT,...

Hi @dsvilarkovic , We used Aliked to extract keypoints. Usually N=1024 is enough. If aiming for better accuracy, you could try N=2048. Higher number will not bring much more improvement,...

Hi @dsvilarkovic , In this case, the inlier_mask can be computed by the something like `(conf_score>thres_1) & (vis_score>thres2)`. The pred_tracks can be constructed using the tracking head of VGGT, the...