Loading...
Loading
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
Neuron registration trends
Score Vision (SN44) is a decentralized computer vision framework on Bittensor that performs Game State Recognition (GSR) for soccer/football video analysis. It combines lightweight validation techniques with distributed mining to reduce costs and processing time for complex video analysis tasks, leveraging frame filtering, keypoint validation, and CLIP-based semantic verification to ensure accuracy while minimizing computational overhead.
Score Vision operates through three coordinated components: (1) Miners receive video streams and perform real-time object detection/tracking on each frame, generating standardized outputs; (2) Validators distribute challenges every 5 minutes, collect miner responses, select keyframes using pitch detection, perform hybrid scoring (keypoint validation via stability/plausibility/reprojection error, semantic CLIP-based verification of object classes), and maintain SQLite database of responses; (3) Subnet Owner manages system health and dynamically adjusts incentive parameters. Miners process video frames locally using NVIDIA GPUs, validators use OpenAI GPT-4o API for frame evaluation, and all participants interact via Bittensor's decentralized network (subnet 44). The lightweight validation approach enables efficient verification without duplicate computation overhead.
GS-HOTA = √(Detection × Association) measures object detection accuracy and tracking consistency. Miner evaluation includes: (1) Quality scoring from detection/tracking accuracy, (2) Consistency measurement across frames, (3) Response time assessment. Validator evaluation includes: (1) Verification accuracy through consensus alignment, (2) Response efficiency in validation process, (3) Overall network contribution.
Rewards are weighted by accuracy and contribution volume. High performers receive additional incentives; poor performance incurs penalties. Frame filtering uses pitch detection, keypoints assessed for stability/plausibility/reprojection error, object detections verified via CLIP-based semantic checks for players, ball, referees, and goalkeepers.
Final scores normalized 0-1 to guide rewards.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
Making every camera intelligent
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository for the latest instructions.
Holder & total-staked metrics as of 1h ago (hourly snapshot)
GPU strongly recommended; CPU-only processing takes ~9138 seconds per challenge (not viable). NVIDIA GPU: ~22 seconds per challenge at 22.36 FPS. Apple Silicon: ~140 seconds (development only). NVMe SSD storage recommended for optimal performance.
No GPU required. Processes challenges every 5 minutes, handles concurrent miner connections, performs response evaluation via OpenAI GPT-4o API, updates weights every 21 minutes. Storage grows ~1.5GB per day. Fast SSD I/O critical for database operations.
Clone the Score Vision repository
git clone https://github.com/score-technologies/score-vision.git && cd score-vision
Create and activate Python virtual environment (Python 3.10+ required)
python -m venv venv && source venv/bin/activate
Install Score Vision miner dependencies
pip install -e .[miner]
Create Bittensor coldkey wallet (if not already created)
python -m bittensor wallet new-coldkey --wallet.name <wallet_name>
Create Bittensor hotkey for miner identity
python -m bittensor wallet new-hotkey --wallet.name <wallet_name> --wallet.hotkey <hotkey_name>
Start the miner node on Bittensor subnet 44
python neurons/miner.py --netuid 44 --subtensor.network finney --wallet.name <wallet_name> --wallet.hotkey <hotkey_name>
Clone the Score Vision repository
git clone https://github.com/score-technologies/score-vision.git && cd score-vision
Create and activate Python virtual environment (Python 3.10+ required)
python -m venv venv && source venv/bin/activate
Install Score Vision validator dependencies
pip install -e .[validator]
Set OpenAI API key (required for GPT-4o evaluations)
export OPENAI_API_KEY='sk-...'
Create Bittensor coldkey wallet (if not already created)
python -m bittensor wallet new-coldkey --wallet.name <wallet_name>
Create Bittensor hotkey for validator identity
python -m bittensor wallet new-hotkey --wallet.name <wallet_name> --wallet.hotkey <hotkey_name>
Start validator using Docker Compose (alternative to manual execution)(optional)
docker-compose up -d
Start the validator node on Bittensor subnet 44
python neurons/validator.py --netuid 44 --subtensor.network finney --wallet.name <wallet_name> --wallet.hotkey <hotkey_name>
