Upload 3 files

Dockerfile

@@ -0,0 +1,19 @@
+# Read the doc: https://huggingface.co/docs/hub/spaces-sdks-docker
+# you will also find guides on how best to write your Dockerfile
+
+FROM python:3.9
+
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+WORKDIR /app
+
+COPY --chown=user ./requirements.txt requirements.txt
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+RUN chmod -R 777 /app
+
+
+COPY --chown=user . /app
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,506 @@
+import os
+import json
+import time
+import asyncio
+import aiohttp
+import zipfile
+import shutil
+from typing import Dict, List, Set, Optional, Tuple
+from urllib.parse import quote
+from datetime import datetime
+from pathlib import Path
+import io
+
+from fastapi import FastAPI, BackgroundTasks, HTTPException, status
+from pydantic import BaseModel, Field
+from huggingface_hub import HfApi, hf_hub_download
+
+# --- Configuration ---
+AUTO_START_INDEX = 20 # Hardcoded default start index if no progress is found
+FLOW_ID = os.getenv("FLOW_ID", "flow_default")
+FLOW_PORT = int(os.getenv("FLOW_PORT", 8001))
+HF_TOKEN = os.getenv("HF_TOKEN", "")
+HF_DATASET_ID = os.getenv("HF_DATASET_ID", "Fred808/BG3") # Source dataset for zip files
+HF_OUTPUT_DATASET_ID = os.getenv("HF_OUTPUT_DATASET_ID", "fred808/helium") # Target dataset for captions
+
+# Progress Tracking File
+PROGRESS_FILE = Path("processing_progress.json")
+# Directory within the HF dataset where the zip files are located
+ZIP_FILE_PREFIX = "frames/"
+
+# Using the full list from the user's original code for actual deployment
+CAPTION_SERVERS = [
+   "https://fred808-pil-4-1.hf.space/analyze",
+    "https://fred808-pil-4-2.hf.space/analyze",
+    "https://fred808-pil-4-3.hf.space/analyze",
+    "https://fred1012-fred1012-gw0j2h.hf.space/analyze",
+    "https://fred1012-fred1012-wqs6c2.hf.space/analyze",
+    "https://fred1012-fred1012-oncray.hf.space/analyze",
+    "https://fred1012-fred1012-4goge7.hf.space/analyze",
+    "https://fred1012-fred1012-z0eh7m.hf.space/analyze",
+    "https://fred1012-fred1012-u95rte.hf.space/analyze",
+    "https://fred1012-fred1012-igje22.hf.space/analyze",
+    "https://fred1012-fred1012-ibkuf8.hf.space/analyze",
+    "https://fred1012-fred1012-nwqthy.hf.space/analyze",
+    "https://fred1012-fred1012-4ldqj4.hf.space/analyze",
+    "https://fred1012-fred1012-pivlzg.hf.space/analyze",
+    "https://fred1012-fred1012-ptlc5u.hf.space/analyze",
+    "https://fred1012-fred1012-u7lh57.hf.space/analyze",
+    "https://fred1012-fred1012-q8djv1.hf.space/analyze",
+    "https://fredalone-fredalone-ozugrp.hf.space/analyze",
+    "https://fredalone-fredalone-9brxj2.hf.space/analyze",
+    "https://fredalone-fredalone-p8vq9a.hf.space/analyze",
+    "https://fredalone-fredalone-vbli2y.hf.space/analyze",
+    "https://fredalone-fredalone-uggger.hf.space/analyze",
+    "https://fredalone-fredalone-nmi7e8.hf.space/analyze",
+    "https://fredalone-fredalone-d1f26d.hf.space/analyze",
+    "https://fredalone-fredalone-461jp2.hf.space/analyze",
+    "https://fredalone-fredalone-3enfg4.hf.space/analyze",
+    "https://fredalone-fredalone-dqdbpv.hf.space/analyze",
+    "https://fredalone-fredalone-ivtjua.hf.space/analyze",
+    "https://fredalone-fredalone-6bezt2.hf.space/analyze",
+    "https://fredalone-fredalone-e0wfnk.hf.space/analyze",
+    "https://fredalone-fredalone-zu2t7j.hf.space/analyze",
+    "https://fredalone-fredalone-dqtv1o.hf.space/analyze",
+    "https://fredalone-fredalone-wclyog.hf.space/analyze",
+    "https://fredalone-fredalone-t27vig.hf.space/analyze",
+    "https://fredalone-fredalone-gahbxh.hf.space/analyze",
+    "https://fredalone-fredalone-kw2po4.hf.space/analyze",
+    "https://fredalone-fredalone-8h285h.hf.space/analyze"
+]
+MODEL_TYPE = "Florence-2-large"
+
+# Temporary storage for images
+TEMP_DIR = Path(f"temp_images_{FLOW_ID}")
+TEMP_DIR.mkdir(exist_ok=True)
+
+# --- Models ---
+class ProcessStartRequest(BaseModel):
+    start_index: int = Field(AUTO_START_INDEX, ge=1, description="The index number of the zip file to start processing from (1-indexed).")
+
+class CaptionServer:
+    def __init__(self, url):
+        self.url = url
+        self.busy = False
+        self.total_processed = 0
+        self.total_time = 0
+        self.model = MODEL_TYPE
+
+    @property
+    def fps(self):
+        return self.total_processed / self.total_time if self.total_time > 0 else 0
+
+# Global state for caption servers
+servers = [CaptionServer(url) for url in CAPTION_SERVERS]
+server_index = 0
+
+# --- Progress Tracking Functions ---
+
+def load_progress() -> Dict:
+    """Loads the processing progress from the JSON file."""
+    if PROGRESS_FILE.exists():
+        try:
+            with PROGRESS_FILE.open('r') as f:
+                return json.load(f)
+        except json.JSONDecodeError:
+            print(f"[{FLOW_ID}] WARNING: Progress file is corrupted. Starting fresh.")
+            # Fall through to return default structure
+    
+    # Default structure
+    return {
+        "last_processed_index": 0,
+        "processed_files": {}, # {index: repo_path}
+        "file_list": [] # Full list of all zip files found in the dataset
+    }
+
+def save_progress(progress_data: Dict):
+    """Saves the processing progress to the JSON file."""
+    try:
+        with PROGRESS_FILE.open('w') as f:
+            json.dump(progress_data, f, indent=4)
+    except Exception as e:
+        print(f"[{FLOW_ID}] CRITICAL ERROR: Could not save progress to {PROGRESS_FILE}: {e}")
+
+# --- Hugging Face Utility Functions ---
+
+async def get_zip_file_list(progress_data: Dict) -> List[str]:
+    """
+    Fetches the list of all zip files from the dataset, or uses the cached list.
+    Updates the progress_data with the file list if a new list is fetched.
+    """
+    if progress_data['file_list']:
+        print(f"[{FLOW_ID}] Using cached file list with {len(progress_data['file_list'])} files.")
+        return progress_data['file_list']
+
+    print(f"[{FLOW_ID}] Fetching full list of zip files from {HF_DATASET_ID}...")
+    try:
+        api = HfApi(token=HF_TOKEN)
+        repo_files = api.list_repo_files(
+            repo_id=HF_DATASET_ID,
+            repo_type="dataset"
+        )
+        
+        # Filter for zip files in the specified directory and sort them alphabetically for consistent indexing
+        zip_files = sorted([
+            f for f in repo_files 
+            if f.startswith(ZIP_FILE_PREFIX) and f.endswith('.zip')
+        ])
+        
+        if not zip_files:
+            raise FileNotFoundError(f"No zip files found in '{ZIP_FILE_PREFIX}' directory of dataset '{HF_DATASET_ID}'.")
+        
+        print(f"[{FLOW_ID}] Found {len(zip_files)} zip files.")
+        
+        # Update and save the progress data
+        progress_data['file_list'] = zip_files
+        save_progress(progress_data)
+        
+        return zip_files
+        
+    except Exception as e:
+        print(f"[{FLOW_ID}] Error fetching file list from Hugging Face: {e}")
+        return []
+
+async def download_and_extract_zip_by_index(file_index: int, repo_file_full_path: str) -> Optional[Path]:
+    """Downloads the zip file for the given index and extracts its contents."""
+    
+    # Extract the base name for the extraction directory
+    zip_full_name = Path(repo_file_full_path).name 
+    course_name = zip_full_name.replace('.zip', '') # Use the file name as the course/job name
+    
+    print(f"[{FLOW_ID}] Processing file #{file_index}: {repo_file_full_path}. Full name: {zip_full_name}")
+    
+    try:
+        # Use hf_hub_download to get the file path
+        zip_path = hf_hub_download(
+            repo_id=HF_DATASET_ID,
+            filename=repo_file_full_path, # Use the full path in the repo
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+        
+        print(f"[{FLOW_ID}] Downloaded to {zip_path}. Extracting...")
+        
+        # Create a temporary directory for extraction
+        extract_dir = TEMP_DIR / course_name
+        # Ensure a clean directory for extraction
+        if extract_dir.exists():
+            shutil.rmtree(extract_dir)
+        extract_dir.mkdir(exist_ok=True)
+        
+        with zipfile.ZipFile(zip_path, 'r') as zip_ref:
+            zip_ref.extractall(extract_dir)
+            
+        print(f"[{FLOW_ID}] Extraction complete to {extract_dir}.")
+        
+        # Clean up the downloaded zip file to save space
+        os.remove(zip_path)
+        
+        return extract_dir
+        
+    except Exception as e:
+        print(f"[{FLOW_ID}] Error downloading or extracting zip for {repo_file_full_path}: {e}")
+        return None
+
+async def upload_captions_to_hf(zip_full_name: str, captions: List[Dict]) -> bool:
+    """Uploads the final captions JSON file to the output dataset."""
+    # Use the full zip name, replacing the extension with .json
+    caption_filename = Path(zip_full_name).with_suffix('.json').name
+    
+    try:
+        print(f"[{FLOW_ID}] Uploading {len(captions)} captions for {zip_full_name} as {caption_filename} to {HF_OUTPUT_DATASET_ID}...")
+        
+        # Create JSON content in memory
+        json_content = json.dumps(captions, indent=2, ensure_ascii=False).encode('utf-8')
+        
+        api = HfApi(token=HF_TOKEN)
+        api.upload_file(
+            path_or_fileobj=io.BytesIO(json_content),
+            path_in_repo=caption_filename,
+            repo_id=HF_OUTPUT_DATASET_ID,
+            repo_type="dataset",
+            commit_message=f"[{FLOW_ID}] Captions for {zip_full_name}"
+        )
+        
+        print(f"[{FLOW_ID}] Successfully uploaded captions for {zip_full_name}.")
+        return True
+        
+    except Exception as e:
+        print(f"[{FLOW_ID}] Error uploading captions for {zip_full_name}: {e}")
+        return False
+
+# --- Core Processing Functions (Modified) ---
+
+async def get_available_server(timeout: float = 300.0) -> CaptionServer:
+    """Round-robin selection of an available caption server."""
+    global server_index
+    start_time = time.time()
+    while True:
+        # Round-robin check for an available server
+        for _ in range(len(servers)):
+            server = servers[server_index]
+            server_index = (server_index + 1) % len(servers)
+            if not server.busy:
+                return server
+        
+        # If all servers are busy, wait for a short period and check again
+        await asyncio.sleep(0.5)
+        
+        # Check if timeout has been reached
+        if time.time() - start_time > timeout:
+            raise TimeoutError(f"Timeout ({timeout}s) waiting for an available caption server.")
+
+async def send_image_for_captioning(image_path: Path, course_name: str, progress_tracker: Dict) -> Optional[Dict]:
+    """Sends a single image to a caption server for processing."""
+    # This function now handles server selection and retries internally
+    MAX_RETRIES = 3
+    for attempt in range(MAX_RETRIES):
+        server = None
+        try:
+            # 1. Get an available server (will wait if all are busy, with a timeout)
+            server = await get_available_server()
+            server.busy = True
+            start_time = time.time()
+            
+            # Print a less verbose message only on the first attempt
+            if attempt == 0:
+                print(f"[{FLOW_ID}] Starting attempt on {image_path.name}...")
+            
+            # 2. Prepare request data
+            form_data = aiohttp.FormData()
+            form_data.add_field('file',
+                                image_path.open('rb'),
+                                filename=image_path.name,
+                                content_type='image/jpeg')
+            form_data.add_field('model_choice', MODEL_TYPE)
+            
+            # 3. Send request
+            async with aiohttp.ClientSession() as session:
+                # Increased timeout to 10 minutes (600s) as requested by user's problem description
+                async with session.post(server.url, data=form_data, timeout=600) as resp:
+                    if resp.status == 200:
+                        result = await resp.json()
+                        caption = result.get("caption")
+                        
+                        if caption:
+                            # Update progress counter
+                            progress_tracker['completed'] += 1
+                            if progress_tracker['completed'] % 50 == 0:
+                                print(f"[{FLOW_ID}] PROGRESS: {progress_tracker['completed']}/{progress_tracker['total']} captions completed.")
+                            
+                            # Log success only if it's not a progress report interval
+                            if progress_tracker['completed'] % 50 != 0:
+                                print(f"[{FLOW_ID}] Success: {image_path.name} captioned by {server.url}")
+                                
+                            return {
+                                "course": course_name,
+                                "image_path": image_path.name,
+                                "caption": caption,
+                                "timestamp": datetime.now().isoformat()
+                            }
+                        else:
+                            print(f"[{FLOW_ID}] Server {server.url} returned success but no caption for {image_path.name}. Retrying...")
+                            continue # Retry with a different server
+                    else:
+                        error_text = await resp.text()
+                        print(f"[{FLOW_ID}] Error from server {server.url} for {image_path.name}: {resp.status} - {error_text}. Retrying...")
+                        continue # Retry with a different server
+                        
+        except (aiohttp.ClientError, asyncio.TimeoutError, TimeoutError) as e:
+            print(f"[{FLOW_ID}] Connection/Timeout error for {image_path.name} on {server.url if server else 'unknown server'}: {e}. Retrying...")
+            continue # Retry with a different server
+        except Exception as e:
+            print(f"[{FLOW_ID}] Unexpected error during captioning for {image_path.name}: {e}. Retrying...")
+            continue # Retry with a different server
+        finally:
+            if server:
+                end_time = time.time()
+                server.busy = False
+                server.total_processed += 1
+                server.total_time += (end_time - start_time)
+
+    print(f"[{FLOW_ID}] FAILED after {MAX_RETRIES} attempts for {image_path.name}.")
+    return None
+
+async def process_dataset_task(start_index: int):
+    """Main task to process the dataset sequentially starting from a given index."""
+    
+    progress = load_progress()
+    file_list = await get_zip_file_list(progress)
+    
+    if not file_list:
+        print(f"[{FLOW_ID}] ERROR: Cannot proceed. File list is empty.")
+        return False
+    
+    # Ensure start_index is within bounds
+    if start_index > len(file_list):
+        print(f"[{FLOW_ID}] WARNING: Start index {start_index} is greater than the total number of files ({len(file_list)}). Exiting.")
+        return True
+    
+    # Determine the actual starting index in the 0-indexed list
+    start_list_index = start_index - 1
+    
+    print(f"[{FLOW_ID}] Starting dataset processing from file index: {start_index} out of {len(file_list)}.")
+    
+    global_success = True
+    
+    for i in range(start_list_index, len(file_list)):
+        file_index = i + 1 # 1-indexed for user display and progress tracking
+        repo_file_full_path = file_list[i]
+        zip_full_name = Path(repo_file_full_path).name 
+        course_name = zip_full_name.replace('.zip', '') # Use the file name as the course/job name
+        
+        # Check if the file has already been successfully processed
+        if str(file_index) in progress['processed_files']:
+            print(f"[{FLOW_ID}] Skipping file #{file_index} ({zip_full_name}): Already processed according to progress file.")
+            progress['last_processed_index'] = file_index
+            save_progress(progress)
+            continue
+            
+        extract_dir = None
+        current_file_success = False
+        
+        try:
+            # 1. Download and Extract
+            extract_dir = await download_and_extract_zip_by_index(file_index, repo_file_full_path)
+            
+            if not extract_dir:
+                raise Exception("Failed to download or extract zip file.")
+            
+            # 2. Find Images
+            # Use recursive glob to find images in subdirectories
+            image_paths = [p for p in extract_dir.glob("**/*") if p.is_file() and p.suffix.lower() in ['.jpg', '.jpeg', '.png']]
+            print(f"[{FLOW_ID}] Found {len(image_paths)} images to process in {zip_full_name}.")
+            
+            if not image_paths:
+                print(f"[{FLOW_ID}] No images found in {zip_full_name}. Marking as complete.")
+                current_file_success = True
+            else:
+                # 3. Process Images (Captioning)
+                progress_tracker = {
+                    'total': len(image_paths),
+                    'completed': 0
+                }
+                print(f"[{FLOW_ID}] Starting captioning for {progress_tracker['total']} images in {zip_full_name}...")
+                
+                # Create a semaphore to limit concurrent tasks to the number of available servers
+                semaphore = asyncio.Semaphore(len(servers))
+                
+                async def limited_send_image_for_captioning(image_path, course_name, progress_tracker):
+                    async with semaphore:
+                        return await send_image_for_captioning(image_path, course_name, progress_tracker)
+                
+                # Create a list of tasks for parallel captioning
+                caption_tasks = [limited_send_image_for_captioning(p, course_name, progress_tracker) for p in image_paths]
+                    
+                # Run all captioning tasks concurrently
+                results = await asyncio.gather(*caption_tasks)
+                
+                # Filter out failed results
+                all_captions = [r for r in results if r is not None]
+                
+                # Final progress report for the current file
+                if len(all_captions) == len(image_paths):
+                    print(f"[{FLOW_ID}] FINAL PROGRESS for {zip_full_name}: Successfully completed all {len(all_captions)} captions.")
+                    current_file_success = True
+                else:
+                    print(f"[{FLOW_ID}] FINAL PROGRESS for {zip_full_name}: Completed with partial result: {len(all_captions)}/{len(image_paths)} captions. Marking as partial failure.")
+                    current_file_success = False
+                
+                # 4. Upload Results
+                if all_captions:
+                    print(f"[{FLOW_ID}] Uploading {len(all_captions)} captions for {zip_full_name}...")
+                    if await upload_captions_to_hf(zip_full_name, all_captions):
+                        print(f"[{FLOW_ID}] Successfully uploaded captions for {zip_full_name}.")
+                        # Partial success in captioning is still a success for the upload step
+                        pass
+                    else:
+                        print(f"[{FLOW_ID}] Failed to upload captions for {zip_full_name}.")
+                        current_file_success = False
+                else:
+                    print(f"[{FLOW_ID}] No captions generated. Skipping upload for {zip_full_name}.")
+                    current_file_success = False
+            
+        except Exception as e:
+            print(f"[{FLOW_ID}] Critical error in process_dataset_task for file #{file_index} ({zip_full_name}): {e}")
+            current_file_success = False
+            global_success = False # Mark overall task as failed if any file fails critically
+            
+        finally:
+            # 5. Cleanup and Update Progress
+            if extract_dir and extract_dir.exists():
+                print(f"[{FLOW_ID}] Cleaned up temporary directory {extract_dir}.")
+                shutil.rmtree(extract_dir, ignore_errors=True)
+                
+            if current_file_success:
+                # Update progress only on successful completion of the file
+                progress['last_processed_index'] = file_index
+                progress['processed_files'][str(file_index)] = repo_file_full_path
+                save_progress(progress)
+                print(f"[{FLOW_ID}] Progress saved: File #{file_index} marked as processed.")
+            else:
+                # If a file fails, we stop the continuous loop to allow for manual intervention or a fresh start
+                print(f"[{FLOW_ID}] File #{file_index} failed. Stopping continuous processing.")
+                global_success = False
+                break
+                
+    print(f"[{FLOW_ID}] All processing loops complete. Overall success: {global_success}")
+    return global_success
+
+# --- FastAPI App and Endpoints ---
+
+app = FastAPI(
+    title=f"Flow Server {FLOW_ID} API",
+    description="Sequentially processes zip files from a dataset, captions images, and tracks progress.",
+    version="1.0.0"
+)
+
+@app.on_event("startup")
+async def startup_event():
+    print(f"Flow Server {FLOW_ID} started on port {FLOW_PORT}.")
+    
+    # Automatically start the processing task
+    progress = load_progress()
+    # Start from the last processed index + 1, or the hardcoded AUTO_START_INDEX if the progress file is new/empty
+    start_index = progress.get('last_processed_index', 0) + 1
+    if start_index < AUTO_START_INDEX:
+        start_index = AUTO_START_INDEX
+
+    # Use a dummy BackgroundTasks object for the startup task
+    # Note: FastAPI's startup events can't directly use BackgroundTasks, but we can use asyncio.create_task
+    # to run the long-running process in the background without blocking the server startup.
+    print(f"[{FLOW_ID}] Auto-starting processing from index: {start_index}...")
+    asyncio.create_task(process_dataset_task(start_index))
+
+@app.get("/")
+async def root():
+    progress = load_progress()
+    return {
+        "flow_id": FLOW_ID,
+        "status": "ready",
+        "last_processed_index": progress['last_processed_index'],
+        "total_files_in_list": len(progress['file_list']),
+        "processed_files_count": len(progress['processed_files']),
+        "total_servers": len(servers),
+        "busy_servers": sum(1 for s in servers if s.busy),
+    }
+
+@app.post("/start_processing")
+async def start_processing(request: ProcessStartRequest, background_tasks: BackgroundTasks):
+    """
+    Starts the sequential processing of zip files from the given index in the background.
+    """
+    start_index = request.start_index
+    
+    print(f"[{FLOW_ID}] Received request to start processing from index: {start_index}. Starting background task.")
+    
+    # Start the heavy processing in a background task so the API call returns immediately
+    # Note: The server is already auto-starting, but this allows for manual restart/override.
+    background_tasks.add_task(process_dataset_task, start_index)
+    
+    return {"status": "processing", "start_index": start_index, "message": "Dataset processing started in background."}
+
+if __name__ == "__main__":
+    import uvicorn
+    # Note: When running in the sandbox, we need to use 0.0.0.0 to expose the port.
+    uvicorn.run(app, host="0.0.0.0", port=FLOW_PORT)

requirements.txt

@@ -0,0 +1,9 @@
+fastapi==0.104.1
+uvicorn==0.24.0
+aiofiles==23.2.1
+python-multipart==0.0.6
+huggingface-hub==0.18.0
+aiohttp
+jinja2
+pydantic
+datasets