probe3d
probe3d copied to clipboard
mbanani
[CVPR 2024] Probing the 3D Awareness of Visual Foundation Models
I am trying to recreate the **point correspondence** between two images from the navi dataset (not just the error evaluation, but the actual image mapping), but there seems to be...
Hello! Thanks for the great work. Are there any plans to release the weights for the depth model trained on Stable Diffusion?
Hello, thank you very much for this excellent work! I am trying to reproduce the paper results on Depth Estimation using NAVI and DINO B16: ``` $ python train_depth.py backbone=dino_b16...
Hi, thank you for the wonderful paper and codebase! I had one clarification question: it looks like there is an extra set of forward passes for the SigLIP ViT blocks...
Hi, thanks for sharing the code! Small question about extracting the clip embedding. Your evals/models/clip.py has a forward method that outputs a tensor of shape: torch.Size([1, 768, 32, 32]) Is...
Hi there, thank you for the amazing work! I noticed while image normalization parameters are passed to the NAVI and NYU datasets (e.g. [here](https://github.com/mbanani/probe3d/blob/c52d00b069d949b2f00c544d4991716df68d5233/evals/datasets/navi.py#L49)) to allow model-specific normalization, in the...
Hello, congrats for the great paper. I am trying to replicate the DIFT results by launching `python train_depth.py backbone=dift +backbone.return_multilayer=True dataset=navi` The model collapses after few iterations, producing constant predictions....
Hello, thank you very much for this excellent work! I am trying to reproduce the paper results on Surface Normal Estimation using NAVI and DINOv2 B14, I using gradient accumulation...
Hello, thank you very much for sharing this great work! I'm in the process of testing other Dino based models on this benchmark. I noticed that for the SPair dataset...
### Summary Switches the interpolation in the DPT head from nearest-neighbor to bilinear. Uses `align_corners=True, consistent with common practice in dense geometric tasks. ### Motivation This change does not produce...
