Johan Edstedt

allow_tf32 can screw up stuff that relies on high precision, don't hardcode. In my case, it messed up torch.cdist so that a bunch of small values went to 0 (not...

Simple inference demo?

3

Hi again, I'm wondering if you could provide a simple inference demo on a single point cloud pair. Right now it seems like I have to use the functions defined...

For me the base script failed to get the triangulations. Probably due to large file size gdrive has some additional checks. Instead of dealing with that stuff just do it...

`@torch.cuda.amp.custom_fwd(cast_inputs=torch.float32)` in https://github.com/cvg/LightGlue/blob/main/lightglue/lightglue.py#L24 gives ``` FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead. @torch.cuda.amp.custom_fwd(cast_inputs=torch.float32) ``` I think it's not as simple as changing it, as I can't find torch.amp.custom_fwd...

"malloc: Incorrect checksum for freed object" when using inputs of different shape in estimate_homography

9

```python import numpy as np import poselib x_A = np.random.randn(9,2) x_B = np.random.randn(4,2) H_est, stats = poselib.estimate_homography(x_A, x_B, {}, {}) ```