Skip to main content

​
Overview

FastAPI backend that serves videos, provides predictions, and handles sorting videos into folders. Includes endpoints for listing videos, managing folders, sorting videos, and triggering model retraining. Location: source/server.py Server: FastAPI with Uvicorn Default Port: 8000

​
Configuration Constants

​
DATA_DIR
Path
default:"data/Favorites/videos"
Directory containing videos and category subfolders.
​
ARTIFACTS_DIR
Path
default:"artifacts"
Directory containing predictions and trained models.
​
INDEX_HTML
Path
default:"index.html"
Path to the frontend HTML file.

​
Pydantic Models

​
SortRequest

Request body schema for sorting a video into a folder. 
class SortRequest(BaseModel):
    filename: str
    folder: str
​
filename
string
required
Video filename (must match pattern \d+\.mp4, e.g., 7234567890123456.mp4)
​
folder
string
required
Target folder name (must exist as a subfolder in DATA_DIR)
Example: 
{
  "filename": "7234567890123456.mp4",
  "folder": "soccer"
}

​
Functions

​
get_folders()

Returns list of category folders with video counts. Returns: list[dict] - List of folders with structure: 
[
  {"name": "soccer", "count": 82},
  {"name": "food", "count": 55},
  {"name": "funny", "count": 5}
]

​
load_predictions()

Startup event handler that loads predictions from artifacts/predictions.json into memory. Populates the global predictions dictionary mapping filename → prediction data.

​
_run_retrain()

Background function that runs the full ML pipeline in sequence:
  1. extract_features.py - Extract features from all videos
  2. train.py - Train classifier on labeled data
  3. predict.py - Generate predictions for unsorted videos
Runs in a separate thread to avoid blocking API requests. Updates retrain_status global state. Timeout: 900 seconds (15 minutes) per script

​
API Endpoints

​
GET /

Serves the main HTML interface. Response: HTML file (index.html) 
curl http://localhost:8000/

​
GET /api/videos

Lists all unsorted videos (in root directory) with their predictions. Response:
​
videos
array
required
Array of unsorted videos with prediction data
​
total
integer
required
Total number of unsorted videos
Video Object Schema:
​
filename
string
required
Video filename
​
predicted_folder
string | null
required
Predicted category folder, or null if no prediction available
​
confidence
float
required
Prediction confidence score (0-1)
​
top_predictions
array
required
Array of top predictions with folder names and confidence scores
Example Response: 
{
  "videos": [
    {
      "filename": "7234567890123456.mp4",
      "predicted_folder": "soccer",
      "confidence": 0.87,
      "top_predictions": [
        {"folder": "soccer", "confidence": 0.87},
        {"folder": "food", "confidence": 0.10}
      ]
    },
    {
      "filename": "7234567890123457.mp4",
      "predicted_folder": null,
      "confidence": 0,
      "top_predictions": []
    }
  ],
  "total": 2
}
cURL Example: 
curl http://localhost:8000/api/videos

​
GET /api/folders

Lists all category folders with video counts. Response:
​
folders
array
required
Array of folder objects
Folder Object Schema:
​
name
string
required
Folder name
​
count
integer
required
Number of videos in this folder
Example Response: 
{
  "folders": [
    {"name": "funny", "count": 5},
    {"name": "food", "count": 55},
    {"name": "soccer", "count": 82}
  ]
}
cURL Example: 
curl http://localhost:8000/api/folders

​
POST /api/sort

Moves a video from root directory to a category folder. Request Body: SortRequest Response:
​
success
boolean
required
Whether the sort operation succeeded
​
filename
string
required
Filename that was sorted
​
folder
string
required
Folder the video was moved to
​
folders
array
required
Updated list of all folders with new counts
Example Response: 
{
  "success": true,
  "filename": "7234567890123456.mp4",
  "folder": "soccer",
  "folders": [
    {"name": "funny", "count": 5},
    {"name": "food", "count": 55},
    {"name": "soccer", "count": 83}
  ]
}
Error Responses:
  • 400 Bad Request - Invalid filename format or folder name
  • 404 Not Found - Video file not found (may already be sorted)
  • 409 Conflict - File already exists in target folder
cURL Example: 
curl -X POST http://localhost:8000/api/sort \
  -H "Content-Type: application/json" \
  -d '{"filename": "7234567890123456.mp4", "folder": "soccer"}'
Security:
  • Validates filename matches pattern \d+\.mp4
  • Prevents path traversal (rejects folders containing .. or /)
  • Checks that destination folder exists and is a directory
  • Validates source file exists and is a file (not already moved)

​
POST /api/retrain

Triggers a full model retraining pipeline in the background. Request Body: None Response:
​
status
string
required
Either "started" if retraining began, or "already_running" if a retrain is in progress
Example Response: 
{
  "status": "started"
}
cURL Example: 
curl -X POST http://localhost:8000/api/retrain
Process:
  1. Checks if retraining is already in progress
  2. Launches background thread to run pipeline
  3. Returns immediately (doesn’t wait for completion)
  4. Pipeline runs: extract_features.py → train.py → predict.py
  5. Reloads predictions into memory when complete

​
GET /api/retrain/status

Checks the status of the retraining pipeline. Response:
​
running
boolean
required
Whether retraining is currently in progress
​
last_result
string | null
required
Result of last retrain attempt:
  • "success" - Completed successfully
  • "Failed at [script]: [error]" - Failed with error message
  • null - No retrain has completed yet
Example Response: 
{
  "running": false,
  "last_result": "success"
}
cURL Example: 
curl http://localhost:8000/api/retrain/status
Polling Pattern: Frontend can poll this endpoint every few seconds while running: true to show progress.

​
GET /videos/

Serves a video file for playback. Path Parameters:
​
filename
string
required
Video filename (must match pattern \d+\.mp4)
Response: Video file with Content-Type: video/mp4 File Lookup:
  1. First checks root directory (unsorted videos)
  2. If not found, searches all category subfolders
  3. Returns 404 if not found anywhere
Example: 
curl http://localhost:8000/videos/7234567890123456.mp4 --output video.mp4
Browser Usage: 
<video src="http://localhost:8000/videos/7234567890123456.mp4" controls></video>
Error Response:
  • 400 Bad Request - Invalid filename format
  • 404 Not Found - Video file not found

​
Running the Server

​
Development

python server.py
Starts Uvicorn server on http://0.0.0.0:8000

​
Production

uvicorn server:app --host 0.0.0.0 --port 8000 --workers 4

​
With Auto-Reload

uvicorn server:app --reload

​
State Management

The server maintains two global state variables:

​
predictions

predictions: dict[str, dict] = {}
Dictionary mapping video filenames to prediction data. Loaded from artifacts/predictions.json at startup and after retraining.

​
retrain_status

retrain_status: dict = {
    "running": False,
    "last_result": None
}
Tracks retraining pipeline status:
  • running: Boolean indicating if retrain is in progress
  • last_result: String with success/failure message from last retrain

​
Dependencies

Required Python packages:
  • fastapi - Web framework
  • uvicorn - ASGI server
  • pydantic - Request/response validation
The server also depends on the ML pipeline scripts:
  • extract_features.py
  • train.py
  • predict.py

​
Architecture Notes

Thread Safety: The retraining pipeline runs in a daemon thread to avoid blocking API requests. Only one retrain can run at a time. File System: All file operations use shutil.move() for atomic moves. Path traversal is prevented through validation. Caching: Predictions are cached in memory and only reloaded after successful retraining. Error Handling: Invalid filenames, missing files, and folder conflicts return appropriate HTTP error codes.

Build docs developers (and LLMs) love

Get started for freeTalk to us