Loading...
Loading
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
Neuron registration trends
Video processing at scale—transcoding, upscaling, and compression—demands expensive GPU infrastructure and coordination across distributed systems. Individual operators cannot efficiently maintain hardware for variable workloads, and buyers lack transparent benchmarks for comparing processing quality and cost across providers.
A decentralized, incentivized network of video processing operators solves this by allowing independent miners to offer upscaling and compression capacity under objective evaluation. Validators issue consistent synthetic challenges (bundled multi-item batches with known ground truth) and score results using standard metrics—VMAF for perceptual quality, compression ratio, and processing latency—creating a public leaderboard of cost-efficient, high-quality performers.
On Bittensor subnet 85, miners run containerized FFmpeg or Video2X services and expose standardized HTTP or Modal serverless endpoints. Validators forward batches of up to five videos per request, download processed outputs, and persist scores in a competitive ranking system. A separate competition mode uses isolated Modal Sandboxes with pinned GitHub repositories, immutable builds, and blinded evaluation to measure encoder quality under controlled constraints—useful for benchmarking new codec implementations or compression strategies against sealed reference datasets.
Use cases were auto-generated from the GitHub repository and may not reflect the latest changes.
Vidaio is a decentralized video processing subnet on Bittensor that enables miners to perform video upscaling and compression tasks, with validators evaluating performance through synthetic and organic workloads. The system supports both traditional inference workflows and competition-based evaluation modes where miners compete to optimize compression quality, speed, and cost efficiency.
The subnet uses a validator-miner architecture where validators query miners with synthetic challenges (bundled multi-item requests) and organic tasks from the organic gateway. Miners run containerized services (Docker Compose) that implement /upscale and /compress routes, forwarding requests either to local HTTP services or Modal serverless workers. Validators forward requests via Bittensor Dendrite, collect responses, download processed outputs, and score them locally using VMAF and custom metrics.
A separate scoring service (port-based microservice) handles batch validation. Organic tasks are coordinated through a Redis queue and separate organic gateway service. Competition mode uses Modal for isolated Sandbox execution with immutable pinned repositories, with results persisted in SQLite and artifacts backed up to S3-compatible storage.
Metagraph snapshots are refreshed on a cadence to track alpha-stake eligibility and enforce deduplication by IP/coldkey.
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.
Synthetic upscaling scoring: VMAF score, PieAPP score, quality score, and length score combine into a final score. Synthetic compression scoring: compression ratio and VMAF metric against target thresholds determine efficiency and quality scores. Cost is tracked as validator-measured GPU/CPU time and estimated from Modal pricing.
Competition mode adds cost-efficiency weighting where cheaper miners score higher relative to quality. Organic tasks are randomly sampled for validation (1 pair for upscaling, up to 5 pairs for compression) and scored similarly. Miners with duplicate IPs or coldkeys are deduplicated and assigned zero scores to prevent Sybil attacks.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
Next-Generation Video Processing Powered By AI
FAQs were auto-generated from the GitHub repository and may not reflect the latest changes.
Holder & total-staked metrics as of 4h ago (hourly snapshot)
NVIDIA drivers and nvidia-container-toolkit required for GPU access. Docker and Docker Compose required for containerized processing services.
Requires Redis for queue management, FFMPEG for video processing, VMAF for quality scoring. Modal authentication required for competition execution.
Install system dependencies
sudo apt update && sudo apt install -y git curl npm python3 python3-venv python3-pip
Clone repository
git clone https://github.com/vidaio-subnet/vidaio-subnet.git && cd vidaio-subnet
Create and activate virtual environment
python3 -m venv venv && source venv/bin/activate
Install package and dependencies
pip install -e .
Configure storage credentials (S3-compatible storage endpoint, access keys, bucket name)
cp miner/.env.template miner/.env && nano miner/.env
Create shared work directory for video processing
mkdir -p /tmp/vidaio-miner-video-tmp && chmod 777 /tmp/vidaio-miner-video-tmp
Start Video2X upscaling service (choose one upscaling option)(optional)
cd miner && docker compose --profile upscaling-video2x up -d upscaling-video2x
Or start FFmpeg upscaling service(optional)
cd miner && docker compose --profile upscaling-ffmpeg up -d upscaling-ffmpeg
Start compression service(optional)
cd miner && docker compose up -d compression
Verify services are healthy(optional)
curl -sf http://localhost:8003/health && curl -sf http://localhost:8004/health
Start miner process
python3 neurons/miner.py --wallet.name [Your_Wallet_Name] --wallet.hotkey [Your_Hotkey_Name] --subtensor.network finney --netuid 85 --axon.port [port] --logging.debug
Run system bootstrap script as root (installs NVIDIA drivers, Docker, Python 3.11)
sudo -E ./bootstrap.sh
Install PM2, Redis, and Node packages
sudo apt update && sudo apt install npm redis-server -y && sudo npm install pm2 -g
Clone repository
git clone https://github.com/vidaio-subnet/vidaio-subnet.git && cd vidaio-subnet
Create and activate virtual environment
python3 -m venv venv && source venv/bin/activate
Install package and dependencies
pip install -e .
Configure environment variables (bucket credentials, Pexels API, W&B API)
cp .env.template .env && nano .env
Install FFmpeg
sudo apt install ffmpeg -y
Build VMAF Docker images for video quality scoring
git clone https://github.com/vidAio-subnet/vmaf.git && cd vmaf && docker build -t vmaf . && docker build -t vmaf_ffmpeg -f Dockerfile.ffmpeg .
Authenticate with Modal for serverless execution
modal setup
Start validator in inference mode
python3 neurons/validator.py --wallet.name [Your_Wallet_Name] --wallet.hotkey [Your_Hotkey_Name] --subtensor.network finney --netuid 85 --axon.port [port] --logging.debug
Start validator in competition mode (after configuring COMPETITION_MODE_ENABLED=true)(optional)
python3 neurons/validator.py --wallet.name [Your_Wallet_Name] --wallet.hotkey [Your_Hotkey_Name] --subtensor.network finney --netuid 85 --validator-mode competition --logging.debug
