Update app.py

app.py

@@ -1,154 +1,162 @@
 import gradio as gr
+import io
 import numpy as np
-import random
-
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
 import torch
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-
-    generator = torch.Generator().manual_seed(seed)
-
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator,
-    ).images[0]
-
-    return image, seed
-
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
+from decord import cpu, VideoReader, bridge
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import BitsAndBytesConfig
+
+# Adjust the model path to Qwen-2B
+MODEL_PATH = "Qwen/Qwen2-VL-2B-Instruct"
+DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
+TORCH_TYPE = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.get_device_capability()[0] >= 8 else torch.float16
+
+DELAY_REASONS = {
+    "Step 1": ["No raw material available", "Person repatching the tire"],
+    "Step 2": ["Person repatching the tire", "Lack of raw material"],
+    "Step 3": ["Person repatching the tire", "Lack of raw material"],
+    "Step 4": ["Person repatching the tire", "Lack of raw material"],
+    "Step 5": ["Person repatching the tire", "Lack of raw material"],
+    "Step 6": ["Person repatching the tire", "Lack of raw material"],
+    "Step 7": ["Person repatching the tire", "Lack of raw material"],
+    "Step 8": ["No person available to collect tire", "Person repatching the tire"]
 }
-"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
 
+def load_video(video_data, strategy='chat'):
+    """Loads and processes video data into a format suitable for model input."""
+    bridge.set_bridge('torch')
+    num_frames = 48
+    
+    if isinstance(video_data, str): 
+        decord_vr = VideoReader(video_data, ctx=cpu(0))
+    else:  
+        decord_vr = VideoReader(io.BytesIO(video_data), ctx=cpu(0))
+    
+    frame_id_list = []
+    total_frames = len(decord_vr)
+    timestamps = [i[0] for i in decord_vr.get_frame_timestamp(np.arange(total_frames))]
+    max_second = round(max(timestamps)) + 1
+    
+    for second in range(max_second):
+        closest_num = min(timestamps, key=lambda x: abs(x - second))
+        index = timestamps.index(closest_num)
+        frame_id_list.append(index)
+        if len(frame_id_list) >= num_frames:
+            break
+
+    video_data = decord_vr.get_batch(frame_id_list)
+    video_data = video_data.permute(3, 0, 1, 2)
+    return video_data
+
+def load_model():
+    """Loads the Qwen-2B model and tokenizer with quantization configurations."""
+    quantization_config = BitsAndBytesConfig(
+        load_in_4bit=True,
+        bnb_4bit_compute_dtype=TORCH_TYPE,
+        bnb_4bit_use_double_quant=True,
+        bnb_4bit_quant_type="nf4"
+    )
+    
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_PATH,
+        torch_dtype=TORCH_TYPE,
+        trust_remote_code=True,
+        quantization_config=quantization_config,
+        device_map="auto"
+    ).eval()
+    
+    return model, tokenizer
+
+def predict(prompt, video_data, temperature, model, tokenizer):
+    """Generates predictions based on the video and textual prompt."""
+    video = load_video(video_data, strategy='chat')
+    
+    inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
+    inputs = {key: value.to(DEVICE) for key, value in inputs.items()}
+    inputs["images"] = [video.to(DEVICE).to(TORCH_TYPE)]
+    
+    gen_kwargs = {
+        "max_new_tokens": 2048,
+        "pad_token_id": tokenizer.pad_token_id,
+        "top_k": 1,
+        "do_sample": False,
+        "top_p": 0.1,
+        "temperature": temperature,
+    }
+    
+    with torch.no_grad():
+        outputs = model.generate(**inputs, **gen_kwargs)
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    return response
+
+def get_analysis_prompt(step_number, possible_reasons):
+    """Constructs the prompt for analyzing delay reasons based on the selected step."""
+    return f"""You are an AI expert system specialized in analyzing manufacturing processes and identifying production delays in tire manufacturing. Your role is to accurately classify delay reasons based on visual evidence from production line footage.
+Task Context:
+You are analyzing video footage from Step {step_number} of a tire manufacturing process where a delay has been detected. Your task is to determine the most likely cause of the delay from the following possible reasons:
+{', '.join(possible_reasons)}
+Required Analysis:
+1. Carefully observe the video for visual cues indicating production interruption.
+2. If no person is visible in any of the frames, the reason probably might be due to his absence
+3. If a person is visible in the video and is observed touching and modifying the layers of the tire, it means there is an issue with the tire being patched, hence they are repatching it.
+4. Compare observed evidence against each possible delay reason.
+5. Select the most likely reason based on visual evidence.
+Please provide your analysis in the following format:
+1. Selected Reason: [State the most likely reason from the given options]
+2. Visual Evidence: [Describe specific visual cues that support your selection]
+3. Reasoning: [Explain why this reason best matches the observed evidence]
+4. Alternative Analysis: [Brief explanation of why other possible reasons are less likely]
+Important: Base your analysis solely on visual evidence from the video. Focus on concrete, observable details rather than assumptions. Clearly state if no person or specific activity is observed."""
+
+# Load model globally
+model, tokenizer = load_model()
+
+def inference(video, step_number):
+    """Analyzes video to predict the most likely cause of delay in the selected manufacturing step."""
+    try:
+        if not video:
+            return "Please upload a video first."
+        
+        possible_reasons = DELAY_REASONS[step_number]
+        prompt = get_analysis_prompt(step_number, possible_reasons)
+        temperature = 0.8
+        response = predict(prompt, video, temperature, model, tokenizer)
+        
+        return response
+    except Exception as e:
+        return f"An error occurred during analysis: {str(e)}"
+
+def create_interface():
+    """Creates the Gradio interface for the Manufacturing Delay Analysis System with examples."""
+    with gr.Blocks() as demo:
+        gr.Markdown("""
+        # Manufacturing Delay Analysis System
+        Upload a video of the manufacturing step and select the step number. 
+        The system will analyze the video and determine the most likely cause of delay.
+        """)
+        
         with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
-            with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
+            with gr.Column():
+                video = gr.Video(label="Upload Manufacturing Video", sources=["upload"])
+                step_number = gr.Dropdown(
+                    choices=list(DELAY_REASONS.keys()),
+                    label="Manufacturing Step"
                 )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
-    )
+                analyze_btn = gr.Button("Analyze Delay", variant="primary")
+            
+            with gr.Column():
+                output = gr.Textbox(label="Analysis Result", lines=10)
+        
+        analyze_btn.click(
+            fn=inference,
+            inputs=[video, step_number],
+            outputs=[output]
+        )
+    
+    return demo
 
 if __name__ == "__main__":
-    demo.launch()
+    demo = create_interface()
+    demo.launch(share=True)