detectron2 VS transformers

I don’t do AI professionally but as a hobby, so this may not be the best way. But the way you described, it seems the user maybe taking the picture a bit further away and there may be other objects in the frame. So you may want to look into some sort of segmentation or have bounding box. This could help the user make sure they are looking at documents for the correct machine.

I think something like detectron2 [1] could help. It is Apache2 license, so commercial friendly. That said the pre-trained weights may not be used for commercial purposes, so you’ll want to check on that.

[1] https://github.com/facebookresearch/detectron2

And not enough true positives either. Add more augmentations in the config. Also make sure the config is set correctly, so that Detectron2 isn't skipping background images: https://github.com/facebookresearch/detectron2/issues/80

I followed the setup of https://github.com/facebookresearch/detectron2/tree/main/projects/TensorMask. But still I can not import it. As I can with from detectron2.projects import point_rend easily from PointRend projects

I have to use this config file with the dataloader which is in https://github.com/facebookresearch/detectron2/blob/main/projects/MViTv2/configs/common/coco_loader.py. I figured that i can use cfg.dataloader.train.dataset.names = "my_dataset_train" for this.

I want to use this config file https://github.com/facebookresearch/detectron2/blob/main/projects/MViTv2/configs/mask_rcnn_mvitv2_t_3x.py like the beneath typical way I use a yaml config file. But giving so many errors one after another that, I even failed to count at this point.

Which is built on https://github.com/facebookresearch/detectron2

Click to Learn more...

Detectron2 (GitHub: https://github.com/facebookresearch/detectron2) is a Facebook AI Research library with state-of-the-art object detection and segmentation algorithms in PyTorch.

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.

In transformers, they tried really hard to have a single function or method to deal with both self and cross attention mechanisms, masking, positional and relative encodings, interpolation etc. While it allows a user to use the same function/method for any model, it has led to severe parameter bloat. Just compare the original implementation of llama by FAIR with the implementation by HF to get an idea.

pip install git+https://github.com/huggingface/transformers.git@main