README.md
| 1 | --- |
| 2 | tags: |
| 3 | - object-detection |
| 4 | - '- vision' |
| 5 | - onnx |
| 6 | license: apache-2.0 |
| 7 | base_model: facebook/detr-resnet-50 |
| 8 | datasets: |
| 9 | - MohamedExperio/ICDAR2019 |
| 10 | --- |
| 11 | |
| 12 | # Model Card for detr-doc-table-detection |
| 13 | |
| 14 | # Model Details |
| 15 | |
| 16 | detr-doc-table-detection is a model trained to detect both **Bordered** and **Borderless** tables in documents, based on [facebook/detr-resnet-50](https://huggingface.co/facebook/detr-resnet-50). |
| 17 | |
| 18 | - **Developed by:** Taha Douaji |
| 19 | - **Shared by [Optional]:** Taha Douaji |
| 20 | - **Model type:** Object Detection |
| 21 | - **Language(s) (NLP):** More information needed |
| 22 | - **License:** More information needed |
| 23 | - **Parent Model:** [facebook/detr-resnet-50](https://huggingface.co/facebook/detr-resnet-50) |
| 24 | - **Resources for more information:** |
| 25 | - [Model Demo Space](https://huggingface.co/spaces/trevbeers/pdf-table-extraction) |
| 26 | - [Associated Paper](https://arxiv.org/abs/2005.12872) |
| 27 | |
| 28 | |
| 29 | |
| 30 | # Uses |
| 31 | |
| 32 | |
| 33 | ## Direct Use |
| 34 | This model can be used for the task of object detection. |
| 35 | |
| 36 | ## Out-of-Scope Use |
| 37 | |
| 38 | The model should not be used to intentionally create hostile or alienating environments for people. |
| 39 | |
| 40 | # Bias, Risks, and Limitations |
| 41 | |
| 42 | |
| 43 | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al. (2021)](https://aclanthology.org/2021.acl-long.330.pdf) and [Bender et al. (2021)](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. |
| 44 | |
| 45 | |
| 46 | |
| 47 | ## Recommendations |
| 48 | |
| 49 | |
| 50 | Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. |
| 51 | |
| 52 | # Training Details |
| 53 | |
| 54 | ## Training Data |
| 55 | |
| 56 | The model was trained on ICDAR2019 Table Dataset |
| 57 | |
| 58 | |
| 59 | # Environmental Impact |
| 60 | |
| 61 | Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). |
| 62 | |
| 63 | |
| 64 | # Citation |
| 65 | |
| 66 | |
| 67 | **BibTeX:** |
| 68 | |
| 69 | |
| 70 | ```bibtex |
| 71 | @article{DBLP:journals/corr/abs-2005-12872, |
| 72 | author = {Nicolas Carion and |
| 73 | Francisco Massa and |
| 74 | Gabriel Synnaeve and |
| 75 | Nicolas Usunier and |
| 76 | Alexander Kirillov and |
| 77 | Sergey Zagoruyko}, |
| 78 | title = {End-to-End Object Detection with Transformers}, |
| 79 | journal = {CoRR}, |
| 80 | volume = {abs/2005.12872}, |
| 81 | year = {2020}, |
| 82 | url = {https://arxiv.org/abs/2005.12872}, |
| 83 | archivePrefix = {arXiv}, |
| 84 | eprint = {2005.12872}, |
| 85 | timestamp = {Thu, 28 May 2020 17:38:09 +0200}, |
| 86 | biburl = {https://dblp.org/rec/journals/corr/abs-2005-12872.bib}, |
| 87 | bibsource = {dblp computer science bibliography, https://dblp.org} |
| 88 | } |
| 89 | ``` |
| 90 | |
| 91 | |
| 92 | # Model Card Authors [optional] |
| 93 | |
| 94 | Taha Douaji in collaboration with Ezi Ozoani and the Hugging Face team |
| 95 | |
| 96 | |
| 97 | # Model Card Contact |
| 98 | |
| 99 | More information needed |
| 100 | |
| 101 | # How to Get Started with the Model |
| 102 | |
| 103 | Use the code below to get started with the model. |
| 104 | |
| 105 | |
| 106 | ```python |
| 107 | from transformers import DetrImageProcessor, DetrForObjectDetection |
| 108 | import torch |
| 109 | from PIL import Image |
| 110 | import requests |
| 111 | |
| 112 | image = Image.open("IMAGE_PATH") |
| 113 | |
| 114 | processor = DetrImageProcessor.from_pretrained("TahaDouaji/detr-doc-table-detection") |
| 115 | model = DetrForObjectDetection.from_pretrained("TahaDouaji/detr-doc-table-detection") |
| 116 | |
| 117 | inputs = processor(images=image, return_tensors="pt") |
| 118 | outputs = model(**inputs) |
| 119 | |
| 120 | # convert outputs (bounding boxes and class logits) to COCO API |
| 121 | # let's only keep detections with score > 0.9 |
| 122 | target_sizes = torch.tensor([image.size[::-1]]) |
| 123 | results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0] |
| 124 | |
| 125 | for score, label, box in zip(results["scores"], results["labels"], results["boxes"]): |
| 126 | box = [round(i, 2) for i in box.tolist()] |
| 127 | print( |
| 128 | f"Detected {model.config.id2label[label.item()]} with confidence " |
| 129 | f"{round(score.item(), 3)} at location {box}" |
| 130 | ) |
| 131 | ``` |