nvdiffrec VS transformers

It depends on which CV research you’re in. In NeRF view synthesis, it’s pretty common to use test sets as validation sets. This has been done in several papers, including oral papers.

I think there came a paper recently nerf2mesh, which I still have to evaluate (but haven't found time yet). There's also https://github.com/NVlabs/nvdiffrec/. And there's cool easy-to-use research software like nerfstudio (at least if you compare it to a lot of the raw code releases from research papers).

For your use case, why is your model devoid of texture? You can try 3D scanning your desired object so that it comes with texture. Either that or use Nvidia's MoMA here to get your object from images.

Have you seen this project: https://github.com/NVlabs/nvdiffrec (I haven't tried it). Also videos tend to have compression. If you can get images you'll get higher quality results with most photogrammetry software. Projects like meshroom are probably better for this if you have high quality pictures. There's a few articles that cover high quality scans that can help also.

You can generate an accurate mesh from a NeRF: https://github.com/NVlabs/nvdiffrec, and measure from that.

NERF specifically generates a radiance field, but there are research codes for turning that into a mesh (https://github.com/NVlabs/nvdiffrec) (not easy to use yet)

Hi, I'm not sure if this is the right place, but I was looking into seeing what the latest photo to model reconstruction looks like from here with NVIDIA (ArXiV paper is included there). There's a couple of neat examples, and after one dumb mistake setup was pretty easy. However, the meshes are not converging except very loosely when using the examples from the paper.

Hi everyone! I'm not sure this is the right place to ask, but I've been drooling over these cool ml and deep learning techniques showcased in videos. I was wondering if anyone could help me out in getting something like nvdiffrec to work with my own sample. https://github.com/NVlabs/nvdiffrec

Is t5x an encoder/decoder architecture?

Some more general options.

The Flax ecosystem

https://github.com/google/flax?tab=readme-ov-file

or dm-haiku

https://github.com/google-deepmind/dm-haiku

were some of the best developed communities in the Jax AI field

Perhaps the “trax” repo? https://github.com/google/trax

Some HF examples https://github.com/huggingface/transformers/tree/main/exampl...

Sadly it seems much of the work is proprietary these days, but one example could be Grok-1, if you customize the details. https://github.com/xai-org/grok-1/blob/main/run.py

The HuggingFace transformers library already has support for a similar method called prompt lookup decoding that uses the existing context to generate an ngram model: https://github.com/huggingface/transformers/issues/27722

I don't think it would be that hard to switch it out for a pretrained ngram model.

Most of this tutorial is based on Hugging Face course about Transformers and on Niels Rogge's Transformers tutorials: make sure to check their work and give them a star on GitHub, if you please ❤️

* Superconvergence + LR range finder + Fast AI's Ranger21 optimizer was the goto optimizer for CNNs, and worked fabulously well, but on transformers, the learning rate range finder sadi 1e-3 was the best, whilst 1e-5 was better. However, the 1 cycle learning rate stuck. https://github.com/huggingface/transformers/issues/16013

Thanks! :) I'm pushing them into transformers, pytorch-gemma and collabing with the Gemma team to resolve all the issues :)

The RoPE fix should already be in transformers 4.38.2: https://github.com/huggingface/transformers/pull/29285

My main PR for transformers which fixes most of the issues (some still left): https://github.com/huggingface/transformers/pull/29402

If you want to tinker with the architecture Hugging Face has a FOSS implementation in transformers: https://github.com/huggingface/transformers/blob/main/src/tr...

If you want to reproduce the training pipeline, you couldn't do that even if you wanted to because you don't have access to thousands of A100s.

I am aware that using mixed precision reduces the stability of weight and there will be little consistency but don't expect it to be this much. I have attached the graph of evaluation metrics. If someone can give me some insight into this issue, that would be great.