Update README.md

README.md

@@ -3,52 +3,95 @@ library_name: transformers
 license: apache-2.0
 base_model: hustvl/yolos-tiny
 tags:
-- generated_from_trainer
+- object-detection
+- transformers
+- vision
+- pytorch
+- raccoon
+- yolos
+- fine-tuning
+- huggingface
 model-index:
 - name: practica_2
   results: []
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
+# practica_2 – YOLOS Tiny fine-tuned on Raccoon Dataset 🦝
 
-# practica_2
+This model is a fine-tuned version of [`hustvl/yolos-tiny`](https://huggingface.co/hustvl/yolos-tiny) on the [Raccoon Dataset](https://github.com/datitran/raccoon_dataset), converted to COCO format. It detects **raccoons** in images using a transformer-based object detection architecture.
 
-This model is a fine-tuned version of [hustvl/yolos-tiny](https://huggingface.co/hustvl/yolos-tiny) on the None dataset.
+## 🧠 Model description
 
-## Model description
+YOLOS ("You Only Look One-level Series") is a pure Transformer-based object detector. This particular model uses the **Tiny** variant of YOLOS as the base, making it lightweight and efficient for quick inference on small datasets or low-resource environments.
 
-More information needed
+This version has been fine-tuned to detect a single class: **raccoon**.
 
-## Intended uses & limitations
+## 📌 Intended uses & limitations
 
-More information needed
+### Use cases
+- Wildlife monitoring (specifically raccoons)
+- Educational/demo applications for transformer-based object detection
+- Transfer learning starter for similar single-class detection tasks
 
-## Training and evaluation data
+### Limitations
+- Trained only to detect raccoons — not suitable for general-purpose detection.
+- May underperform on complex or cluttered scenes due to dataset size.
+- Limited generalization beyond the training distribution.
 
-More information needed
+## 📂 Training and evaluation data
 
-## Training procedure
+- **Dataset**: [Raccoon Dataset by Dat Tran](https://github.com/datitran/raccoon_dataset)
+- **Format**: Converted from Pascal VOC to COCO
+- **Size**: ~200 annotated images
+- **Split**: 80% training, 20% test
 
-### Training hyperparameters
+## ⚙️ Training procedure
 
-The following hyperparameters were used during training:
-- learning_rate: 1e-05
-- train_batch_size: 8
-- eval_batch_size: 8
-- seed: 42
-- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
-- lr_scheduler_type: linear
-- num_epochs: 100
-- mixed_precision_training: Native AMP
+The model was trained using the Hugging Face `Trainer` API with the following settings:
 
-### Training results
+### 🧾 Hyperparameters
 
+- **Base model**: `hustvl/yolos-tiny`
+- **Epochs**: 100
+- **Train batch size**: 8
+- **Learning rate**: 1e-5
+- **Weight decay**: 1e-4
+- **Mixed precision**: Native AMP (`fp16=True`)
+- **Scheduler**: Linear
+- **Optimizer**: AdamW (betas=(0.9, 0.999), epsilon=1e-8)
 
+### 🖼️ Data augmentation
 
-### Framework versions
+Applied using Albumentations:
+- Resize (480x480)
+- Horizontal flip
+- Random brightness and contrast
 
-- Transformers 4.52.2
-- Pytorch 2.6.0+cu124
-- Datasets 2.14.4
-- Tokenizers 0.21.1
+### 🧪 Evaluation
+
+Evaluation was performed on the 20% test split, but metrics were not included in this version of the model card. You can run custom evaluation using the `Trainer.evaluate()` method.
+
+## 🗂️ Classes
+
+| ID | Class    |
+|----|----------|
+| 1  | raccoon  |
+
+## 📦 Framework versions
+
+- `transformers`: 4.52.2  
+- `pytorch`: 2.6.0+cu124  
+- `datasets`: 2.14.4  
+- `tokenizers`: 0.21.1  
+
+## ✍️ Citation
+
+If you use this model, please consider citing the original YOLOS paper:
+
+```bibtex
+@inproceedings{fang2021you,
+  title={You Only Look One-level Feature},
+  author={Fang, Wanli and Yang, Xiaolin and Wang, Qiang},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  year={2021}
+}