From 0c55d47cded0c711a4359afb04d73ffb86b7d913 Mon Sep 17 00:00:00 2001 From: NielsRogge <48327001+NielsRogge@users.noreply.github.com> Date: Tue, 22 Mar 2022 08:51:13 +0100 Subject: [PATCH] Add GLPN (#16199) * First draft * Fix logits calculation * Improve tests * Add copied from statements * Fix base_model_prefix * Improve implementation, upload new models * Update design * Fix integration test * Add model to README and toctree * Add document image * Apply suggestions from code review * Apply suggestions from code review Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * Add decoder_hidden_size attribute * Update design of decoder * Add DepthEstimatorOutput class * Rename in_index to head_in_index and add feature extractor tests * Apply suggestions from code review * Apply suggestions from code review * Update pretrained model name and add to doc tests * Remove test.py script * Update copied from statements and clean up Co-authored-by: Niels Rogge Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> --- README.md | 1 + README_ko.md | 1 + README_zh-hans.md | 1 + README_zh-hant.md | 1 + docs/source/_toctree.yml | 2 + docs/source/index.mdx | 2 + docs/source/model_doc/glpn.mdx | 60 ++ src/transformers/__init__.py | 19 + src/transformers/modeling_outputs.py | 30 + src/transformers/models/__init__.py | 1 + .../models/auto/configuration_auto.py | 3 + src/transformers/models/auto/modeling_auto.py | 1 + src/transformers/models/glpn/__init__.py | 60 ++ .../models/glpn/configuration_glpn.py | 137 ++++ .../models/glpn/convert_glpn_to_pytorch.py | 219 +++++ .../models/glpn/feature_extraction_glpn.py | 140 ++++ src/transformers/models/glpn/modeling_glpn.py | 758 ++++++++++++++++++ src/transformers/utils/dummy_pt_objects.py | 24 + .../utils/dummy_vision_objects.py | 7 + tests/glpn/__init__.py | 0 tests/glpn/test_feature_extraction_glpn.py | 127 +++ tests/glpn/test_modeling_glpn.py | 356 ++++++++ utils/check_repo.py | 1 + utils/documentation_tests.txt | 13 + 24 files changed, 1964 insertions(+) create mode 100644 docs/source/model_doc/glpn.mdx create mode 100644 src/transformers/models/glpn/__init__.py create mode 100644 src/transformers/models/glpn/configuration_glpn.py create mode 100644 src/transformers/models/glpn/convert_glpn_to_pytorch.py create mode 100644 src/transformers/models/glpn/feature_extraction_glpn.py create mode 100755 src/transformers/models/glpn/modeling_glpn.py create mode 100644 tests/glpn/__init__.py create mode 100644 tests/glpn/test_feature_extraction_glpn.py create mode 100644 tests/glpn/test_modeling_glpn.py diff --git a/README.md b/README.md index ee5fa1fe25..c044565b20 100644 --- a/README.md +++ b/README.md @@ -265,6 +265,7 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab. 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon. 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le. +1. **[GLPN](https://huggingface.co/docs/transformers/master/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim. 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever. 1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**. 1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki. diff --git a/README_ko.md b/README_ko.md index 36bf28697c..8be562ffc5 100644 --- a/README_ko.md +++ b/README_ko.md @@ -244,6 +244,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab. 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon. 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le. +1. **[GLPN](https://huggingface.co/docs/transformers/master/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim. 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever. 1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy. 1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**. diff --git a/README_zh-hans.md b/README_zh-hans.md index 2de4ca361b..fef5cdc09b 100644 --- a/README_zh-hans.md +++ b/README_zh-hans.md @@ -268,6 +268,7 @@ conda install -c huggingface transformers 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (来自 CNRS) 伴随论文 [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) 由 Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab 发布。 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (来自 Google Research) 伴随论文 [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) 由 James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon 发布。 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (来自 CMU/Google Brain) 伴随论文 [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) 由 Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le 发布。 +1. **[GLPN](https://huggingface.co/docs/transformers/master/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。 1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (来自 EleutherAI) 随仓库 [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) 发布。作者为 Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy 发布。 1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (来自 OpenAI) 伴随论文 [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) 由 Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever** 发布。 diff --git a/README_zh-hant.md b/README_zh-hant.md index dcf4a1b21c..92c94ffd5e 100644 --- a/README_zh-hant.md +++ b/README_zh-hant.md @@ -280,6 +280,7 @@ conda install -c huggingface transformers 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab. 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon. 1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le. +1. **[GLPN](https://huggingface.co/docs/transformers/master/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim. 1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever. 1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy. 1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**. diff --git a/docs/source/_toctree.yml b/docs/source/_toctree.yml index c2da3ac559..60bb5030d9 100644 --- a/docs/source/_toctree.yml +++ b/docs/source/_toctree.yml @@ -218,6 +218,8 @@ title: FSMT - local: model_doc/funnel title: Funnel Transformer + - local: model_doc/glpn + title: GLPN - local: model_doc/herbert title: HerBERT - local: model_doc/ibert diff --git a/docs/source/index.mdx b/docs/source/index.mdx index 51f6adbbcd..c99a4cb3b2 100644 --- a/docs/source/index.mdx +++ b/docs/source/index.mdx @@ -89,6 +89,7 @@ conversion utilities for the following models. 1. **[FlauBERT](model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab. 1. **[FNet](model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon. 1. **[Funnel Transformer](model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le. +1. **[GLPN](model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim. 1. **[GPT](model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever. 1. **[GPT-2](model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**. 1. **[GPT-J](model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki. @@ -200,6 +201,7 @@ Flax), PyTorch, and/or TensorFlow. | FlauBERT | ✅ | ❌ | ✅ | ✅ | ❌ | | FNet | ✅ | ✅ | ✅ | ❌ | ❌ | | Funnel Transformer | ✅ | ✅ | ✅ | ✅ | ❌ | +| GLPN | ❌ | ❌ | ✅ | ❌ | ❌ | | GPT Neo | ❌ | ❌ | ✅ | ❌ | ✅ | | GPT-J | ❌ | ❌ | ✅ | ❌ | ✅ | | Hubert | ❌ | ❌ | ✅ | ✅ | ❌ | diff --git a/docs/source/model_doc/glpn.mdx b/docs/source/model_doc/glpn.mdx new file mode 100644 index 0000000000..30e9277d76 --- /dev/null +++ b/docs/source/model_doc/glpn.mdx @@ -0,0 +1,60 @@ + + +# GLPN + + + +This is a recently introduced model so the API hasn't been tested extensively. There may be some bugs or slight +breaking changes to fix it in the future. If you see something strange, file a [Github Issue](https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title). + + + +## Overview + +The GLPN model was proposed in [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim. +GLPN combines [SegFormer](segformer)'s hierarchical mix-Transformer with a lightweight decoder for monocular depth estimation. The proposed decoder shows better performance than the previously proposed decoders, with considerably +less computational complexity. + +The abstract from the paper is the following: + +*Depth estimation from a single image is an important task that can be applied to various fields in computer vision, and has grown rapidly with the development of convolutional neural networks. In this paper, we propose a novel structure and training strategy for monocular depth estimation to further improve the prediction accuracy of the network. We deploy a hierarchical transformer encoder to capture and convey the global context, and design a lightweight yet powerful decoder to generate an estimated depth map while considering local connectivity. By constructing connected paths between multi-scale local features and the global decoding stream with our proposed selective feature fusion module, the network can integrate both representations and recover fine details. In addition, the proposed decoder shows better performance than the previously proposed decoders, with considerably less computational complexity. Furthermore, we improve the depth-specific augmentation method by utilizing an important observation in depth estimation to enhance the model. Our network achieves state-of-the-art performance over the challenging depth dataset NYU Depth V2. Extensive experiments have been conducted to validate and show the effectiveness of the proposed approach. Finally, our model shows better generalisation ability and robustness than other comparative models.* + +Tips: + +- One can use [`GLPNFeatureExtractor`] to prepare images for the model. + + + + Summary of the approach. Taken from the original paper. + +This model was contributed by [niels](