yifeihu commited on
Commit
26b6542
1 Parent(s): b66970a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +22 -22
README.md CHANGED
@@ -15,18 +15,17 @@ TF-ID (Table/Figure IDentifier) is a family of object detection models finetuned
15
  | Model | Model size | Model Description |
16
  | ------- | ------------- | ------------- |
17
  | TF-ID-base[[HF]](https://huggingface.co/yifeihu/TF-ID-base) | 0.23B | Extract tables/figures and their caption text
18
- | TF-ID-large[[HF]](https://huggingface.co/yifeihu/TF-ID-large) | 0.77B | Extract tables/figures and their caption text
19
  | TF-ID-base-no-caption[[HF]](https://huggingface.co/yifeihu/TF-ID-base-no-caption) | 0.23B | Extract tables/figures without caption text
20
- | TF-ID-large-no-caption[[HF]](https://huggingface.co/yifeihu/TF-ID-large-no-caption) | 0.77B | Extract tables/figures without caption text
21
  All TF-ID models are finetuned from [microsoft/Florence-2](https://huggingface.co/microsoft/Florence-2-large-ft) checkpoints.
22
 
23
- The models were finetuned with papers from Hugging Face Daily Papers. All bounding boxes are manually annotated and checked by humans.
 
 
 
24
 
25
- TF-ID models take an image of a single paper page as the input, and return bounding boxes for all tables and figures in the given page.
26
-
27
- TF-ID-base and TF-ID-large draw bounding boxes around tables/figures and their caption text.
28
-
29
- TF-ID-base-no-caption and TF-ID-large-no-caption draw bounding boxes around tables/figures without their caption text.
30
 
31
  ![image/png](https://huggingface.co/yifeihu/TF-ID-base/resolve/main/td-id-caption.png)
32
 
@@ -34,6 +33,10 @@ Object Detection results format:
34
  {'\<OD>': {'bboxes': [[x1, y1, x2, y2], ...],
35
  'labels': ['label1', 'label2', ...]} }
36
 
 
 
 
 
37
  ## Benchmarks
38
 
39
  We tested the models on paper pages outside the training dataset. The papers are a subset of huggingface daily paper.
@@ -59,18 +62,16 @@ Use the code below to get started with the model.
59
  ```python
60
  import requests
61
  from PIL import Image
62
- from transformers import AutoProcessor, AutoModelForCausalLM
63
 
64
- model = AutoModelForCausalLM.from_pretrained("yifeihu/TF-ID-large", trust_remote_code=True)
65
- processor = AutoProcessor.from_pretrained("yifeihu/TF-ID-large", trust_remote_code=True)
66
 
67
  prompt = "<OD>"
68
-
69
  url = "https://huggingface.co/yifeihu/TF-ID-base/resolve/main/arxiv_2305_10853_5.png?download=true"
70
  image = Image.open(requests.get(url, stream=True).raw)
71
 
72
  inputs = processor(text=prompt, images=image, return_tensors="pt")
73
-
74
  generated_ids = model.generate(
75
  input_ids=inputs["input_ids"],
76
  pixel_values=inputs["pixel_values"],
@@ -78,8 +79,8 @@ generated_ids = model.generate(
78
  do_sample=False,
79
  num_beams=3
80
  )
81
- generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
82
 
 
83
  parsed_answer = processor.post_process_generation(generated_text, task="<OD>", image_size=(image.width, image.height))
84
 
85
  print(parsed_answer)
@@ -87,16 +88,15 @@ print(parsed_answer)
87
 
88
  To visualize the results, see [this tutorial notebook](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-florence-2-on-detection-dataset.ipynb) for more details.
89
 
90
- ## Finetuning Code and Dataset
91
-
92
- Coming soon!
93
-
94
  ## BibTex and citation info
95
 
96
  ```
97
- @misc{TF-ID,
98
- url={[https://huggingface.co/yifeihu/TF-ID-base](https://huggingface.co/yifeihu/TF-ID-base)},
99
- title={TF-ID: Table/Figure IDentifier for academic papers},
100
- author={"Yifei Hu"}
 
 
 
101
  }
102
  ```
 
15
  | Model | Model size | Model Description |
16
  | ------- | ------------- | ------------- |
17
  | TF-ID-base[[HF]](https://huggingface.co/yifeihu/TF-ID-base) | 0.23B | Extract tables/figures and their caption text
18
+ | TF-ID-large[[HF]](https://huggingface.co/yifeihu/TF-ID-large) (Recommended) | 0.77B | Extract tables/figures and their caption text
19
  | TF-ID-base-no-caption[[HF]](https://huggingface.co/yifeihu/TF-ID-base-no-caption) | 0.23B | Extract tables/figures without caption text
20
+ | TF-ID-large-no-caption[[HF]](https://huggingface.co/yifeihu/TF-ID-large-no-caption) (Recommended) | 0.77B | Extract tables/figures without caption text
21
  All TF-ID models are finetuned from [microsoft/Florence-2](https://huggingface.co/microsoft/Florence-2-large-ft) checkpoints.
22
 
23
+ - The models were finetuned with papers from Hugging Face Daily Papers. All bounding boxes are manually annotated and checked by humans.
24
+ - TF-ID models take an image of a single paper page as the input, and return bounding boxes for all tables and figures in the given page.
25
+ - TF-ID-base and TF-ID-large draw bounding boxes around tables/figures and their caption text.
26
+ - TF-ID-base-no-caption and TF-ID-large-no-caption draw bounding boxes around tables/figures without their caption text.
27
 
28
+ **Large models are always recommended!**
 
 
 
 
29
 
30
  ![image/png](https://huggingface.co/yifeihu/TF-ID-base/resolve/main/td-id-caption.png)
31
 
 
33
  {'\<OD>': {'bboxes': [[x1, y1, x2, y2], ...],
34
  'labels': ['label1', 'label2', ...]} }
35
 
36
+ ## Training Code and Dataset
37
+ - Dataset: [yifeihu/TF-ID-arxiv-papers](https://huggingface.co/datasets/yifeihu/TF-ID-arxiv-papers)
38
+ - Code: [github.com/ai8hyf/TF-ID](https://github.com/ai8hyf/TF-ID)
39
+
40
  ## Benchmarks
41
 
42
  We tested the models on paper pages outside the training dataset. The papers are a subset of huggingface daily paper.
 
62
  ```python
63
  import requests
64
  from PIL import Image
65
+ from transformers import AutoProcessor, AutoModelForCausalLM
66
 
67
+ model = AutoModelForCausalLM.from_pretrained("yifeihu/TF-ID-base", trust_remote_code=True)
68
+ processor = AutoProcessor.from_pretrained("yifeihu/TF-ID-base", trust_remote_code=True)
69
 
70
  prompt = "<OD>"
 
71
  url = "https://huggingface.co/yifeihu/TF-ID-base/resolve/main/arxiv_2305_10853_5.png?download=true"
72
  image = Image.open(requests.get(url, stream=True).raw)
73
 
74
  inputs = processor(text=prompt, images=image, return_tensors="pt")
 
75
  generated_ids = model.generate(
76
  input_ids=inputs["input_ids"],
77
  pixel_values=inputs["pixel_values"],
 
79
  do_sample=False,
80
  num_beams=3
81
  )
 
82
 
83
+ generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
84
  parsed_answer = processor.post_process_generation(generated_text, task="<OD>", image_size=(image.width, image.height))
85
 
86
  print(parsed_answer)
 
88
 
89
  To visualize the results, see [this tutorial notebook](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-florence-2-on-detection-dataset.ipynb) for more details.
90
 
 
 
 
 
91
  ## BibTex and citation info
92
 
93
  ```
94
+ @misc{TF-ID,
95
+ author = {Yifei Hu},
96
+ title = {TF-ID: Table/Figure IDentifier for academic papers},
97
+ year = {2024},
98
+ publisher = {GitHub},
99
+ journal = {GitHub repository},
100
+ howpublished = {\url{https://github.com/ai8hyf/TF-ID}},
101
  }
102
  ```