README.md

---
license: apache-2.0
---

# This is a YOLOX model trained on the DOCLAYNET dataset


## Training:
- It has been trained using the repo: `https://github.com/Megvii-BaseDetection/YOLOX/tree/main/`
- The doclaynet dataset was converted to COCO format internally
- The model was trained on the dataset for 300 epochs

Here's the Loss function (over 300 epochs or 600k iterations)
![image.png](https://cdn-uploads.huggingface.co/production/uploads/6335e808d184e6b53c49da1b/_Xa-keI6uGWedwqx7NudV.png)


## Quant performance

- Also - the model compared to the yolox-l.pt that was released by unstructured under no defined license on the doclaynet dataset is captured here

(On the test dataset)
![image.png](https://cdn-uploads.huggingface.co/production/uploads/6335e808d184e6b53c49da1b/0kXhBn-vw9Awjuvjs7hhu.png)

(On the validation dataset)
![image.png](https://cdn-uploads.huggingface.co/production/uploads/6335e808d184e6b53c49da1b/5y6lb36bUUuUMzyM-Ucfn.png)