Workpam commited on
Commit
60dcb20
·
verified ·
1 Parent(s): 2da178a

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +33 -32
app.py CHANGED
@@ -1,40 +1,41 @@
 
1
  import gradio as gr
2
- from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
3
- import torch
4
- from PIL import Image
5
  import requests
 
 
6
 
7
- # Load model & processor
8
- model_name = "nlpconnect/vit-gpt2-image-captioning"
9
- model = VisionEncoderDecoderModel.from_pretrained(model_name)
10
- feature_extractor = ViTImageProcessor.from_pretrained(model_name)
11
- tokenizer = AutoTokenizer.from_pretrained(model_name)
12
 
13
- # Use GPU if available
14
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
15
- model.to(device)
16
-
17
- # Caption generation function
18
  def generate_alt_text(image):
19
- if image is None:
20
- return "No image uploaded"
21
-
22
- image = image.convert("RGB")
23
- pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values
24
- pixel_values = pixel_values.to(device)
 
 
 
 
 
 
 
 
 
 
25
 
26
- output_ids = model.generate(pixel_values, max_length=64, num_beams=4)
27
- caption = tokenizer.decode(output_ids[0], skip_special_tokens=True)
28
- return caption.strip().capitalize()
 
 
29
 
30
- # Gradio UI
31
- iface = gr.Interface(
32
  fn=generate_alt_text,
33
- inputs=gr.Image(type="pil", label="Upload an Image"),
34
- outputs=gr.Textbox(label="Generated Alt Text"),
35
- title="Image Alt Text Generator",
36
- description="Upload an image and get a detailed, AI-generated alt text using Hugging Face models"
37
- )
38
-
39
- # Launch app
40
- iface.launch()
 
1
+ import os
2
  import gradio as gr
 
 
 
3
  import requests
4
+ import base64
5
+ import io
6
 
7
+ HF_MODEL = "llava-hf/llava-v1.6-mistral-7b-hf"
8
+ HF_TOKEN = os.getenv("HF_API_TOKEN")
9
+ API_URL = f"https://api-inference.huggingface.co/models/{HF_MODEL}"
10
+ HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"}
 
11
 
 
 
 
 
 
12
  def generate_alt_text(image):
13
+ buffered = io.BytesIO()
14
+ image.save(buffered, format="JPEG")
15
+ b64 = base64.b64encode(buffered.getvalue()).decode()
16
+
17
+ payload = {
18
+ "inputs": [
19
+ {
20
+ "role": "user",
21
+ "content": [
22
+ {"type": "image", "image": b64},
23
+ {"type": "text", "text": "Describe this image in detail"}
24
+ ],
25
+ }
26
+ ],
27
+ "options": {"wait_for_model": True},
28
+ }
29
 
30
+ resp = requests.post(API_URL, headers=HEADERS, json=payload)
31
+ if resp.status_code != 200:
32
+ return f"Error: {resp.status_code}"
33
+ output = resp.json()
34
+ return output[0]["generated_text"].strip()
35
 
36
+ gr.Interface(
 
37
  fn=generate_alt_text,
38
+ inputs=gr.Image(type="pil"),
39
+ outputs=gr.Textbox(label="Alt Text from LLaVA"),
40
+ title="Inference‑based Rich Alt Text"
41
+ ).launch()