# Vintern-1B-v3.5 – Stage 1 (Findings)

## 1. Model Overview

This model is part of a Vision-Language AI system designed for chest X-ray analysis in Vietnamese clinical settings.

The full pipeline consists of 3 stages:
- Stage 1: Findings generation (image → radiology findings)
- Stage 2: Impression generation (findings → clinical impression)
- Stage 3: Multi-turn conversation (findings + impression + dialogue)

This repository corresponds to:

- Stage: 1 (Findings)
- Task: Generate radiology findings from chest X-ray images
- Domain: Vietnamese medical imaging (Chest X-ray)

The model is fine-tuned from a vision-language backbone and evaluated against multiple architectures (InternVL, Vintern, Qwen2-VL, MiniCPM-V, LaVy).

Among all models, Qwen2-VL-7B achieved the best performance, but this model is provided for benchmarking and comparison.

---

## 2. Installation

```bath
pip install torch torchvision transformers decord pillow
```

## 3. Inference
GPU with bfloat16 is recommended.
```python
import numpy as np
import torch
import torchvision.transforms as T
from decord import VideoReader, cpu
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([
        T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=MEAN, std=STD)
    ])
    return transform

def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio

def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1)
        for i in range(1, n + 1)
        for j in range(1, n + 1)
        if i * j <= max_num and i * j >= min_num
    )
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        split_img = resized_img.crop(box)
        processed_images.append(split_img)

    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)

    return processed_images

def load_image(image_file, input_size=448, max_num=12):
    image = Image.open(image_file).convert('RGB')
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values

path = "THP2903/Vintern-1B-v3_5_finding_v2"

model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True
).eval().cuda()

tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False)

pixel_values = load_image("your_image.jpg", max_num=12).to(torch.bfloat16).cuda()

generation_config = dict(max_new_tokens=1024, do_sample=True)

question = "<image>\nẢnh chụp xray benh nhân nam, 48 tuổi PA có dấu hieu gi?"
response = model.chat(tokenizer, pixel_values, question, generation_config)

print(response)
```

## 4. Notes
Input must be a chest X-ray image
Output is radiology findings (not final diagnosis)
This model uses the original inference pipeline without modification
For best performance, consider using Qwen2-VL-7B