** About the Role ** We're looking for engineers to build and optimize video compression systems. You'll develop compression pipelines that reduce file sizes while maintaining quality, working with AV1 encoding and quality metrics. ** Responsibilities ** - Design and implement video compression pipelines using AV1 codec - Optimize compression algorithms to achieve high compression ratios (target: 5-20x) while meeting VMAF quality thresholds - Implement quality control mechanisms to ensure compressed videos meet specified VMAF thresholds (typically 85-93) - (Further task) Build scene detection and classification systems to adapt encoding parameters based on content type (animation, gaming, text, etc.) - (Further task) Optimize encoding performance for real-time processing of video chunks ** Technical Requirements ** * Video Input Requirements: * - Primary: 4K resolution videos (3840x2160) - Secondary: HD resolution videos (1920x1080) - Solution must handle both resolutions efficiently within the time constraints * Required Skills: * - Strong experience with video codecs, especially AV1 (SVT-AV1 and AV1-nvenc preferred) - Proficiency in FFmpeg and video processing libraries - Experience with VMAF and video quality assessment metrics - Python programming skills - Understanding of video encoding parameters (CQ, bitrate, profiles, etc.) - Knowledge of video containers and formats (MP4, Y4M, etc.) * Preferred Skills: * - Machine learning/AI experience, particularly for scene classification - Experience with PyTorch or similar deep learning frameworks - Understanding of computer vision for video analysis ** Technical Specifications ** * Encoding Requirements * - Codec: AV1(required) - Container: MP4 - Profile : Main - Pixel Format: yuv420p - Sample Aspect Ratio: 1:1 - Preferred Encoder : SVT-AV1 or AV1-nvenc * Quality Metrics: * - Primary: VMAF (Video Multimethod Assessment Fusion) - Must meet or exceed specified VMAF thresholds (85-93 range) - Compression rate must be < 0.80 (minimum 1.25x compression required) * Performance Expectations: * - Achieve compression ratios of 5-20x while maintaining quality - Process video chunks efficiently (10-30 second segments) - Successfully meet VMAF thresholds in 90%+ of cases * Hardware Constraints: * - Solutions must perform well on reasonable hardware configurations - Baseline performance target: RTX 4090 GPU (or equivalent) and CPU with 3GHz clock speed - Your implementation should be optimized to meet timing requirements on this hardware tier - Hardware choice is flexible, but performance must meet the 30-second processing requirement on the baseline configuration ** What You'll Build ** Your solution must deliver compressed videos that: - Meet or exceed the specified VMAF quality threshold (typically 85-93) - Achieve high compression ratios (target: 5-20x, compression_rate < 0.20) - Complete processing within 20 ~ 30 seconds per video chunk (30s length 4K video) The implementation approach is flexible—use scene classification, adaptive encoding, quality prediction, or other methods. The focus is on reliably meeting the VMAF threshold with strong compression within the time limit. ** Performance & Evaluation ** * Scoring Criteria: * - Compression Rate (70% weight): Ratio of compressed to original file size - lower is better - VMAF Quality Score (30% weight): Quality assessment metric - must meet or exceed threshold - Final score combines both metrics with quality thresholds * Success Metrics: * - Compression ratio: 5-20x (compression_rate < 0.20) - VMAF score: Meets or exceeds specified threshold (85-93 range) - Processing time: Complete pipeline within 30 seconds - Success rate: 90%+ of videos meet all requirements ** Nice to Have ** - Experience with GPU-accelerated video encoding (NVENC, etc.) - Knowledge of VMAF or other perceptual quality metrics - Experience with video upscaling or other video processing tasks - Contributions to open-source video processing projects - Understanding of parallel processing and optimization techniques ** Applicants must provide: ** - Examples of previous video compression or AV1 work - Relevant GitHub or portfolio links - Brief explanation of how they’d approach the compression-speed tradeoff ** Deliverables: ** - Complete encoding pipeline (scripts + configs) - Performance benchmarks on provided test videos - VMAF evaluation results - Documentation on how to run the pipeline We will provide sample 4K and 1080p videos for benchmark consistency. ________________________________________________________________________________________________________ Note: This role requires strong optimization skills and the ability to balance compression efficiency, quality preservation, and processing speed. Candidates should be prepared to demonstrate their ability to meet the 30-second processing requirement while achieving high compression ratios and quality scores, handling both 4K and HD video resolutions.