ColorVideoVDP — color spatiotemporal visible difference predictor (Cambridge Rainbow Group)
Project page for ColorVideoVDP, a full-reference perceptual video quality metric modeling chromatic and spatiotemporal contrast sensitivity, outputting JOD scores plus distortion maps and "distograms"; differentiable so it can be used as a training loss.
Link
Related resources
- ColorVideoVDP — reference implementationPyTorch implementation of the ColorVideoVDP full-reference metric (SIGGRAPH 2024), usable as a CLI quality tool or as a differentiable perc…
- UVQ — Universal Video Quality (Google)Google's no-reference perceptual video quality model for user-generated content, scoring compression, content and distortion without a pris…
- DOVER — Disentangled Objective Video Quality EvaluatorICCV 2023 no-reference VQA model that outputs separate aesthetic and technical quality scores, with ONNX export and a DOVER-Mobile variant…
- BAND-2k — banding artifact database and detectorDataset of 2,000 banding images plus a dual-CNN banding evaluator that vastly outperforms CAMBI on banding-quality correlation (SRCC 0.88 v…
- QTGMC — AviSynth Wiki ReferenceCanonical parameter reference for QTGMC, the highest-quality motion-compensated deinterlacer, including double-rate output and FPSDivisor s…
- hdr10plus_tool — Codec Wiki Usage GuideCommunity reference with flags and FFmpeg pipe recipes for extracting, injecting and inspecting HDR10+ dynamic metadata in HEVC/MKV.
- PyNvVideoCodec API Programming GuideNVIDIA's Python bindings over the Video Codec SDK for GPU-accelerated decode/encode pipelines feeding ML and analytics workloads.
- FFmpeg 8.0 libplacebo Filter — Full Option ReferenceComplete searchable option reference for FFmpeg's libplacebo filter, including the frame-analysis compute shader path for static-HDR10 and…
- libplacebo Options ReferenceAuthoritative option reference for libplacebo's GPU video rendering: BT.2390 tone mapping, gamut mapping, contrast recovery and LUT sizing.
- COVER — Comprehensive Video Quality EvaluatorCVPRW 2024 winner of the AIS 2024 video quality assessment challenge; tri-branch Swin/ConvNet/CLIP architecture doing real-time 1080p no-re…
- VapourSynth Python ReferenceCanonical scripting API reference for VapourSynth's frameserver, covering core.std, resize and text filter namespaces used in high-end enco…
- MLT Framework — MLT XML DocumentationReference for authoring MLT XML producer/consumer graphs, the declarative render format behind Kdenlive, Shotcut and Flowblade.