|
|
--- |
|
|
datasets: |
|
|
- hongrui/mimic_chest_xray_v_1 |
|
|
--- |
|
|
# 🩺 CheXNet-MedScan-Report-Gen |
|
|
|
|
|
**CheXNet-MedScan-Report-Gen** is an image captioning model for generating diagnostic text reports from chest X-ray images. It combines the power of a pretrained CheXNet encoder (based on DenseNet121) and a bidirectional LSTM decoder to produce sequence-based textual descriptions. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Model Architecture |
|
|
|
|
|
- **Encoder:** DenseNet121 (CheXNet) with classifier removed |
|
|
- **Decoder:** Bidirectional LSTM with dropout |
|
|
- **Feature dimension:** 1024 |
|
|
- **Embedding dimension:** 256 |
|
|
- **Hidden dimension:** 512 |
|
|
- **Vocabulary size:** 5000 |
|
|
- **Dropout:** 0.5 |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔧 Usage |
|
|
|
|
|
You can load the model using the Hugging Face Transformers library: |
|
|
|
|
|
```python |
|
|
from transformers import AutoModel, AutoConfig |
|
|
|
|
|
config = AutoConfig.from_pretrained("khalednabawi11/Chexnet-MedScan-Report-Gen", trust_remote_code=True) |
|
|
model = AutoModel.from_pretrained("khalednabawi11/Chexnet-MedScan-Report-Gen", config=config, trust_remote_code=True) |
