vidaio / docs /incentive_mechanism.md

Update model files

f72237c verified 5 months ago

preview code

raw

history blame contribute delete

29.9 kB

VIDAIO Subnet Validation & Incentive Mechanism

Overview
Task Types
- Upscaling Tasks
- Compression Tasks
Quality Validation Metrics
- VMAF (Video Multi-Method Assessment Fusion)
- PIE-APP (Perceptual Image-Error Assessment)
Upscaling System
Compression System
- Compression Scoring
- Compression Penalty & Bonus System
Implementation Guidelines
Technical Specifications
Mathematical Properties

Overview

The VIDAIO subnet validation mechanism ensures quality and reliability of miners' contributions through comprehensive assessment systems for different video processing tasks. The mechanism evaluates video processing performance using industry-standard VMAF and PIE-APP metrics, combined with advanced scoring systems that reward consistency and penalize poor performance.

Key Features:

🎯 Multi-task support for upscaling and compression operations
📊 Quality validation using industry-standard VMAF and PIE-APP metrics
🏆 Performance-based incentive systems with exponential rewards
📈 Historical performance tracking with rolling 10-round windows
⚖️ Balanced penalty/bonus multipliers encouraging sustained excellence
📦 Dynamic content processing (5s to 320s capability)

Task Types

Upscaling Tasks

Upscaling tasks require miners to enhance video quality by increasing resolution while maintaining or improving visual fidelity. These tasks focus on quality improvement as the primary objective.

Key Characteristics:

Primary Goal: Quality enhancement and resolution improvement
Quality Metrics: PIE-APP for scoring, VMAF for threshold validation
Content Length: Dynamic processing durations (5s to 320s)
Scoring Focus: Quality improvement with content length consideration

Compression Tasks

Compression tasks require miners to reduce video file sizes while maintaining quality above specified VMAF thresholds. These tasks focus on efficiency optimization balancing file size reduction with quality preservation.

Key Characteristics:

Primary Goal: File size reduction with quality maintenance
Quality Metrics: VMAF for both scoring and threshold validation
Compression Rate: File size reduction efficiency measurement
Scoring Focus: Compression efficiency with quality threshold compliance

Quality Validation Metrics

VMAF (Video Multi-Method Assessment Fusion)

VMAF serves as the foundational video quality assessment metric, comparing frame-by-frame quality between original and processed videos. This metric provides objective measurement of subjective video quality as perceived by humans.

Key Characteristics

Purpose: Frame-by-frame quality comparison
Range: 0-100 (higher values indicate better quality)
Usage: Threshold validation and quality scoring
Industry Standard: Widely adopted in professional video processing

Mathematical Implementation

Harmonic Mean Calculation:

H = n / (1/S_1 + 1/S_2 + ... + 1/S_n)

Where:

S_i: VMAF score for frame i (i = 1, 2, ..., n)
n: Total number of frames in the video
H: Harmonic mean emphasizing poor-quality frame impact

Why Harmonic Mean?

The harmonic mean approach provides several critical advantages:

Advantage	Description	Impact
Sensitivity to Low Values	Heavily penalizes poor-quality frames	Ensures consistent quality
Quality Consistency	Prevents miners from neglecting frame quality	Maintains processing standards
Threshold Function	Validates authentic processing processes	Prevents gaming attempts

Note: VMAF scores are calculated using 5 random frames for both upscaling and compression tasks.

PIE-APP (Perceptual Image-Error Assessment through Pairwise Preference)

PIE-APP provides deep learning-based perceptual similarity assessment between original and processed video frames, serving as the primary quality scoring mechanism for upscaling tasks.

Technical Specifications

Parameter	Value	Description
Scale Range	(-∞, ∞)	Theoretical range
Practical Range	0 to 5+	Positive values (lower = better)
Processing Interval	Every frame	Default frame sampling rate
Implementation	Deep learning-based	Advanced perceptual assessment

Calculation Process

Step 1: Raw PIE-APP Score

PIE-APP_score = (Σ abs(d(F_i, F'_i))) / n

Where:

F_i: Frame i from original video
F'_i: Corresponding frame i from processed video
d(F_i, F'_i): Perceptual difference between frames
n: Number of processed frames (4 random frames)

Step 2: Score Normalization

1. Cap values: max(Average_PIE-APP, 2.0)
2. Sigmoid normalization: normalized_score = 1/(1+exp(-Average_PIE-APP))
3. Final transformation: Convert "lower is better" to "higher is better" (0-1 range)

Visual Score Transformation

The PIE-APP scoring system uses sophisticated mathematical transformations:

Sigmoid normalization function for PIE-APP scores !
Final score transformation converting to 0-1 range !

Upscaling System

Upscaling Scoring

Quality Score Calculation

VMAF Threshold Validation:

If VMAF_score < VMAF_threshold:
    S_Q = 0 (Zero score for quality violation)
Else:
    S_Q = PIE-APP_Final_Score

Where:

S_Q: Quality score for upscaling
VMAF_score: Achieved VMAF quality score
VMAF_threshold: Minimum required VMAF score
PIE-APP_Final_Score: Normalized PIE-APP score (0-1 range)

Metric Integration

Metric	Range	Primary Function	Usage
VMAF	0-100	Threshold validation	Upscaling verification
PIE-APP	0-1 (final)	Quality scoring	Performance evaluation

Content Length Scoring

Dynamic Content Length Requests

Miners actively request content processing durations within 35-second evaluation windows, enabling optimized resource allocation and performance assessment.

Available Processing Durations

Duration	Status	Availability
5s	✅ Default	Currently Available
10s	✅ Available	Currently Available
20s	🔄 Coming Soon	Future Release
40s	🔄 Coming Soon	Future Release
80s	🔄 Coming Soon	Future Release
160s	🔄 Coming Soon	Future Release
320s	🔄 Coming Soon	Future Release

Current Limitation: Processing durations up to 10 seconds are currently supported.

Length Score Mathematical Model

Formula:

S_L = log(1 + content_length) / log(1 + 320)

Parameters:

content_length: Processing duration in seconds
S_L: Normalized length score (0 to 1)

Performance Analysis Table

Duration (s)	S_L Score	Percentage	Improvement	Performance Tier
5	0.3105	31.05%	Baseline	Default
10	0.4155	41.55%	+33%	Significant Gain
20	0.5275	52.75%	+27%	Strong Performance
40	0.6434	64.34%	+22%	High Capability
80	0.7614	76.14%	+18%	Advanced Processing
160	0.8804	88.04%	+16%	Expert Level
320	1.0000	100.00%	+14%	Maximum Score

Logarithmic Scaling Benefits

Benefit	Description	Impact
Fair Distribution	Balanced scoring across duration ranges	Equitable competition
Diminishing Returns	Reduced gains for extreme durations	Prevents over-optimization
Normalized Output	Consistent 0-1 scoring range	Standardized evaluation
Capacity Recognition	Rewards longer processing capabilities	Incentivizes advancement

Strategic Insights:

Optimal Entry Point: 10s processing provides largest relative improvement (+33%)
Scaling Pattern: Each duration doubling yields progressively smaller benefits
Maximum Achievement: 320s processing represents theoretical performance ceiling

Upscaling Final Score

Score Component Architecture

The comprehensive upscaling scoring system integrates two fundamental metrics:

Component	Symbol	Description	Weight
Quality Score	S_Q	Processing accuracy and output quality	W1 = 0.5
Length Score	S_L	Content processing capacity	W2 = 0.5

Preliminary Score Calculation

Formula:

S_pre = S_Q × W1 + S_L × W2

Current Configuration:

W1 = 0.5 (Quality weight)
W2 = 0.5 (Length weight)

Dynamic Adjustment: Weights are continuously optimized based on real-world performance data and network requirements.

Final Score Transformation

The preliminary score undergoes exponential transformation for enhanced performance differentiation:

Formula:

S_F = 0.1 × e^(6.979 × (S_pre - 0.5))

Parameters:

S_F: Final upscaling score
S_pre: Preliminary combined score
e: Euler's number (≈2.718)

Performance Tier Analysis

S_pre	S_F Score	Multiplier	Performance Tier	Reward Category
0.30	0.0248	0.25×	Poor Performance	Significant Penalty
0.36	0.0376	0.38×	Below Average	Moderate Penalty
0.42	0.0572	0.57×	Low Average	Minor Penalty
0.48	0.0870	0.87×	Near Average	Slight Penalty
0.54	0.1322	1.32×	Above Average	Moderate Reward
0.60	0.2010	2.01×	Good Performance	Strong Reward
0.66	0.3055	3.05×	High Performance	Major Reward
0.72	0.4643	4.64×	Very High Performance	Excellent Reward
0.78	0.7058	7.06×	Excellent Performance	Outstanding Reward
0.84	1.0728	10.73×	Outstanding Performance	Elite Reward
0.90	1.6307	16.31×	Elite Performance	Maximum Reward

Exponential Function Characteristics

System Benefits:

Feature	Description	Strategic Impact
Enhanced Differentiation	Clear performance tier separation	Competitive advantage clarity
Reward Amplification	16× multiplier difference (top vs bottom)	Strong performance incentives
Competitive Optimization	Non-linear improvement rewards	Encourages continuous advancement
Exponential Scaling	Small S_pre gains yield large S_F improvements	High-performance focus

Strategic Performance Guidelines:

Minimum Target: Achieve S_pre > 0.6 for meaningful reward activation
Optimization Focus: Exponential curve creates powerful excellence incentives
High-Performance Strategy: Small quality improvements at elevated levels yield disproportionate benefits

Graph Analysis

! Length Score Analysis

Upscaling Penalty & Bonus System

Historical Performance Multiplier Architecture

The advanced upscaling scoring system incorporates a rolling 10-round historical performance window to evaluate consistency patterns and apply dynamic multipliers based on sustained performance trends.

System Formula

Final Adjusted Score = S_F × Performance Multiplier
Performance Multiplier = Bonus Multiplier × S_F Penalty × S_Q Penalty

Bonus System (Excellence Rewards)

Activation Criteria: S_F > 0.32 in mining round

Mathematical Model:

bonus_multiplier = 1.0 + (bonus_count / 10) × 0.15

System Characteristics:

Parameter	Value	Description
Maximum Bonus	+15%	All 10 rounds achieve S_F > 0.32
Scaling Method	Linear	Based on consistency frequency
Primary Purpose	Sustained excellence reward	Long-term performance incentive

Example Calculation: 7/10 rounds with S_F > 0.32 → 1.105× multiplier (+10.5% bonus)

S_F Penalty System (Performance Penalties)

Activation Criteria: S_F < 0.20 in mining round

Mathematical Model:

penalty_f_multiplier = 1.0 - (penalty_f_count / 10) × 0.20

System Characteristics:

Parameter	Value	Description
Maximum Penalty	-20%	All 10 rounds achieve S_F < 0.20
Scaling Method	Linear	Based on poor performance frequency
Primary Purpose	Performance consistency enforcement	Discourages sustained poor results

Example Calculation: 4/10 rounds with S_F < 0.20 → 0.92× multiplier (-8% penalty)

S_Q Penalty System (Quality Penalties)

Activation Criteria: S_Q < 0.25 in mining round

Mathematical Model:

penalty_q_multiplier = 1.0 - (penalty_q_count / 10) × 0.25

System Characteristics:

Parameter	Value	Description
Maximum Penalty	-25%	All 10 rounds achieve S_Q < 0.25
Scaling Method	Linear	Based on quality failure frequency
Primary Purpose	Quality standard enforcement	Strongest penalty (quality is critical)

Example Calculation: 3/10 rounds with S_Q < 0.25 → 0.925× multiplier (-7.5% penalty)

Performance Multiplier Case Studies

Miner Category	Avg S_F	Avg S_Q	Bonus Rate	S_F Penalty	S_Q Penalty	Final Multiplier	Net Effect
Elite Miner	0.854	0.717	10/10	0/10	0/10	1.150×	+15.0%
Good Miner	0.653	0.571	0/10	0/10	0/10	1.000×	±0.0%
Average Miner	0.511	0.505	0/10	0/10	3/10	0.925×	-7.5%
Poor Miner	0.311	0.411	0/10	10/10	10/10	0.600×	-40.0%

Penalty Analysis

! Length Score Analysis

System Benefits & Strategic Impact

Core System Benefits:

Benefit	Description	Strategic Impact
🎯 Consistency Rewards	Elite miners maintain sustained +15% bonus	Long-term competitive advantage
⚡ Responsive Penalties	Poor performance accumulates immediate penalties	Rapid feedback mechanism
🔄 Recovery Incentive	Miners can improve multipliers over 10 rounds	Encourages continuous improvement
⚖️ Balanced Impact	Quality penalties are strongest (-25% max)	Emphasizes quality importance
📈 Progressive Scaling	Linear scaling prevents extreme swings	Maintains system stability

Compression System

Compression Scoring

Task Parameters

Each compression task includes specific requirements:

Parameter	Description	Range	Impact
VMAF Threshold	Minimum acceptable quality score	0-100	Quality validation
Original File Size	Baseline for compression calculation	Variable	Compression rate baseline
Target Optimization	Balance between compression and quality	Dynamic	Performance strategy

Compression Rate Calculation

Formula:

C = compressed_file_size / original_file_size

Where:

C: Compression rate (0 < C ≤ 1)
compressed_file_size: Size of processed video file
original_file_size: Size of original video file

Characteristics:

Lower C values indicate better compression (smaller files)
C = 1.0 means no compression achieved
C < 1.0 indicates successful compression

VMAF Quality Assessment

Implementation:

VMAF_score = Harmonic_Mean(VMAF_frame_1, VMAF_frame_2, ..., VMAF_frame_n)

Where:

VMAF_frame_i: VMAF score for frame i
n: Number of sampled frames (5 random frames)
Harmonic_Mean: Emphasizes poor-quality frame impact

Final Compression Score Calculation

Threshold Validation:

If VMAF_score < VMAF_threshold:
    S_f = 0 (Zero score for quality violation)
Else:
    S_f = w_c × (1 - C^1.5) + w_vmaf × (VMAF_score - VMAF_threshold) / (100 - VMAF_threshold)

Parameters:

S_f: Final compression score
w_c: Weight for compression rate (default: 0.8)
w_vmaf: Weight for VMAF score (default: 0.2)
C: Compression rate
VMAF_score: Achieved VMAF quality score
VMAF_threshold: Minimum required VMAF score

Mathematical Properties

Compression Rate Component:

Formula: w_c × (1 - C^1.5)
Range: [0, w_c] (0 when C = 1, w_c when C = 0)
Curve: Concave function emphasizing compression efficiency
Exponent 1.5: Provides balanced reward for compression achievements

VMAF Quality Component:

Formula: w_vmaf × (VMAF_score - VMAF_threshold) / (100 - VMAF_threshold)
Range: [0, w_vmaf] (0 at threshold, w_vmaf at maximum quality)
Normalization: Scales quality improvement relative to achievable range
Linear scaling: Direct correlation between quality improvement and score

Graph Analysis

! Scoring impact

Performance Analysis Examples

Scenario	C	VMAF_score	VMAF_threshold	S_f	Performance Tier
Excellent	0.3	85	70	0.669 + 0.100 = 0.769	Outstanding
Good	0.5	80	70	0.517 + 0.067 = 0.584	Strong
Average	0.7	75	70	0.331 + 0.033 = 0.364	Acceptable
Poor Quality	0.4	65	70	0.000	Failed
No Compression	1.0	90	70	0.000 + 0.133 = 0.133	Inefficient

Strategic Guidelines

Miner Strategy	Focus Area	Target Metrics	Expected Outcome
Quality-First	Maintain high VMAF scores	VMAF_score >> VMAF_threshold	Consistent moderate scores
Compression-First	Maximize file size reduction	C << 1.0	Variable scores based on quality
Balanced Approach	Optimize both factors	Moderate C + Good VMAF	Optimal long-term performance
Threshold Gaming	Minimal quality compliance	VMAF_score ≈ VMAF_threshold	Low scores, high risk

Compression Penalty & Bonus System

Historical Performance Multiplier Architecture

The compression scoring system incorporates a rolling 10-round historical performance window to evaluate consistency patterns and apply dynamic multipliers based on sustained performance trends.

System Formula

Final Adjusted Compression Score = S_f × Performance Multiplier
Performance Multiplier = Bonus Multiplier × S_f Penalty × VMAF Penalty

Bonus System (Excellence Rewards)

Activation Criteria: S_f > 0.74 in compression mining round

Mathematical Model:

bonus_multiplier = 1.0 + (bonus_count / 10) × 0.15

System Characteristics:

Parameter	Value	Description
Maximum Bonus	+15%	All 10 rounds achieve S_f > 0.74
Scaling Method	Linear	Based on consistency frequency
Primary Purpose	Sustained excellence reward	Long-term performance incentive

Example Calculation: 7/10 rounds with S_f > 0.74 → 1.105× multiplier (+10.5% bonus)

S_f Penalty System (Performance Penalties)

Activation Criteria: S_f < 0.4 in compression mining round

Mathematical Model:

penalty_f_multiplier = 1.0 - (penalty_f_count / 10) × 0.20

System Characteristics:

Parameter	Value	Description
Maximum Penalty	-20%	All 10 rounds achieve S_f < 0.4
Scaling Method	Linear	Based on poor performance frequency
Primary Purpose	Performance consistency enforcement	Discourages sustained poor results

Example Calculation: 4/10 rounds with S_f < 0.4 → 0.92× multiplier (-8% penalty)

Compression Performance Multiplier Case Studies

| Miner Category | Avg S_f | Avg VMAF Margin | Bonus Rate | S_f Penalty | Final Multiplier | Net Effect | |----------------|---------|-----------------|------------|-------------|---------------|---------------------|----------------| | Elite Compressor | 0.654 | +15 | 10/10 | 0/10 | 1.150× | +15.0% | | Good Compressor | 0.453 | +8 | 0/10 | 0/10 | 1.000× | ±0.0% | | Average Compressor | 0.311 | +3 | 0/10 | 0/10 | 0.910× | -9.0% | | Poor Compressor | 0.211 | -2 | 0/10 | 10/10 | 0.490× | -51.0% |

Compression Penalty Analysis

Strategic Impact:

Quality-first approach strongly incentivized through higher penalties
Consistency rewards for miners maintaining quality above threshold
Immediate feedback through zero-score for quality violations
Balanced optimization encouraged through dual-factor scoring

Implementation Guidelines

Performance Monitoring

Real-time Operations:

✅ Scores calculated in real-time during mining operations
✅ Historical performance data maintained for trend analysis and multiplier calculation
✅ Weight adjustments implemented based on network-wide performance metrics
✅ Performance multipliers updated after each mining round

Scoring Strategy Recommendations

Upscaling Performance-Based Guidelines:

Miner Category	Primary Focus	Strategic Recommendations
New Miners	Foundation Building	Focus on achieving consistent S_pre > 0.5 before optimizing for length
Established Miners	Quality Optimization	Prioritize quality improvements when S_pre > 0.6 to avoid S_Q penalties
Elite Miners	Consistency Maintenance	Maintain consistency above S_F > 0.32 to secure maximum bonus multipliers
Recovery Phase	Systematic Improvement	Focus on quality (S_Q > 0.25) first, then performance (S_F > 0.20) to restore multipliers

Compression Performance-Based Guidelines:

Miner Category	Primary Focus	Strategic Recommendations
New Compressors	Quality Compliance	Focus on maintaining VMAF_score > VMAF_threshold + 5
Established Compressors	Balanced Optimization	Optimize both compression rate and quality maintenance
Elite Compressors	Consistency Excellence	Maintain S_f > 0.74 consistently for bonus multipliers
Recovery Phase	Quality Restoration	Focus on VMAF compliance first, then compression optimization

Future Enhancement Roadmap

Planned Developments:

Extended Content Length Support - Processing durations up to 320s
Dynamic Weight Adjustment Algorithms - Automated optimization based on network performance
Advanced Quality Metrics Integration - Additional assessment parameters
Multi-dimensional Scoring Parameters - Enhanced evaluation criteria
Adaptive Difficulty Scaling - Network performance-based adjustments
Advanced Penalty/Bonus Optimization - Network-wide performance distribution analysis
Seasonal Performance Multiplier Adjustments - Time-based optimization cycles
Cross-Task Performance Integration - Unified scoring across upscaling and compression

Technical Specifications

Core System Parameters

Parameter	Current Value	Configurable Range	Implementation Notes
Default Content Length	5s	5s - 10s	Actively configurable by miners
Quality Weight (W1)	0.5	0.0 - 1.0	Dynamically adjusted based on network data
Length Weight (W2)	0.5	0.0 - 1.0	Dynamically adjusted based on network data

Compression System Parameters

Parameter	Current Value	Configurable Range	Implementation Notes
Compression Rate Weight (w_c)	0.8	0.6 - 0.9	Balances compression efficiency vs quality
VMAF Score Weight (w_vmaf)	0.2	0.1 - 0.4	Balances quality maintenance vs compression
Compression Rate Exponent	1.5	1.2 - 2.0	Controls compression reward curve steepness
VMAF Safety Margin	+5	+3 - +10	Quality buffer above threshold
Zero-Score Threshold	VMAF_score < VMAF_threshold	Fixed	Immediate penalty for quality violations