Upload folder using huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
metrics:
|
| 6 |
+
- cer
|
| 7 |
+
- wer
|
| 8 |
+
base_model:
|
| 9 |
+
- facebook/deit-base-patch16-224
|
| 10 |
+
- ai4bharat/IndicBART
|
| 11 |
+
tags:
|
| 12 |
+
- scene-text-recognition
|
| 13 |
+
- text-recognition
|
| 14 |
+
- computer-vision
|
| 15 |
+
- language-model
|
| 16 |
+
---
|
| 17 |
+
# trocr-indic
|
| 18 |
+
|
| 19 |
+
This model utilizes the trocr approach to predict the **Indic Texts** from **cropped_images**.
|
| 20 |
+
## Model Details
|
| 21 |
+
|
| 22 |
+
The model follows the TrOCR approach of training OCR for Scene Texts. Since, there is scarcity for generalized model for majority of Indian Languages, this model serves it replacement.
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
*Courtesty: TrOCR - [original paper](https://huggingface.co/papers/2109.10282)*
|
| 26 |
+
|
| 27 |
+
The model is trained for the following languages:
|
| 28 |
+
|
| 29 |
+
- Assamese
|
| 30 |
+
- Bengali
|
| 31 |
+
- Gujarati
|
| 32 |
+
- Hindi
|
| 33 |
+
- Kannada
|
| 34 |
+
- Malayalam
|
| 35 |
+
- Marathi
|
| 36 |
+
- Odia
|
| 37 |
+
- Punjabi
|
| 38 |
+
- Telugu
|
| 39 |
+
- Tamil
|
| 40 |
+
|
| 41 |
+
### Model Description
|
| 42 |
+
|
| 43 |
+
**IMPORTANT**
|
| 44 |
+
Although the model is trained on these languages due to limitations of IndicBART, the model is trained with only Devnagiri Scripts.
|
| 45 |
+
|
| 46 |
+
The output is in the following format:
|
| 47 |
+
```
|
| 48 |
+
<LANGUAGE TOKEN> <TEXT TOKENS> <EOS TOKEN>
|
| 49 |
+
```
|
| 50 |
+
|
| 51 |
+
The following flowchart gives a better picture on the approach of training and inference regarding this model.
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
- **Datasets used:** [IndicSTR12](https://cvit.iiit.ac.in/research/projects/cvit-projects/indicstr)
|
| 57 |
+
- **Developed by:** Aarya Devarla
|
| 58 |
+
- **Model type:** Visio-Lingual Model / Vision-Language Model
|
| 59 |
+
- **License:** mit
|
| 60 |
+
- **Finetuned from model:** deit, indicBART
|
| 61 |
+
|
| 62 |
+
### Results
|
| 63 |
+
|
| 64 |
+
| Metric | Assamese | Bengali | Gujarati | Hindi | Kannada | Malayalam | Marathi | Odia | Punjabi | Tamil | Telugu |
|
| 65 |
+
|--------|----------|---------|----------|-------|---------|-----------|---------|------|---------|-------|--------|
|
| 66 |
+
| CER | 0.069 | 0.133 | 0.058 | 0.075 | 0.212 | 0.154 | 0.082 | 0.120 | 0.097 | 0.122 | 0.220 |
|
| 67 |
+
| WER | 0.205 | 0.395 | 0.192 | 0.283 | 0.576 | 0.519 | 0.312 | 0.375 | 0.304 | 0.409 | 0.612 |
|
| 68 |
+
|
| 69 |
+
Well, the model isn't perfect. But it's a start.
|
| 70 |
+
|
| 71 |
+
## Limitations
|
| 72 |
+
|
| 73 |
+
The main limitation comes from IndicBART which is primarily trained on IndicTexts.
|
| 74 |
+
|
| 75 |
+
### Recommendations
|
| 76 |
+
|
| 77 |
+
Since the TrOCR is modular in approach one can just swap out the IndicBART model and train it with new model. Must keep in mind about the preprocessing and outputs.
|