Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Commit History
-
docs: add CPU-only note to hero and uniformize polarity phrasing Running
CarolinePascalUnverified signatureMention measurements are CPU-only in the shared page hero (RGB + Depth), and standardize About-page metric descriptions from "Lower/Higher = …" to "Lower/Higher is …" across both benchmarks.3f8d017
-
fix: populate encoder/decoder columns and keep Decoder header readable
CarolinePascalUnverified signatureAdd encoder/decoder to the RGB/depth keep_keys so the values survive the row projection instead of rendering blank, and floor the results column width at the longest header word so a single-word header like "Decoder" no longer clips.6a455a0 -
docs: rewrite column, parameter, and about-page descriptions
CarolinePascalUnverified signatureReword the column tooltips, parameter-group blurbs, and the RGB/depth About pages for clarity and consistency, and expand the depth About page with access-pattern, reproduce-locally, and contribute sections.ff929d2 -
feat: clarify submit-form codecs and align depth g/crf options
CarolinePascalUnverified signatureShow friendly "AV1 (libsvtav1/libaom-av1)" labels in the submit codec picker while keeping raw ffmpeg values, and reuse the RGB g/crf submit options for depth so both benchmarks expose the same sweep range.88f8401 -
feat: add hidden-by-default encoder/decoder columns
CarolinePascalUnverified signatureSurface the encoder and decoder libraries next to the codec in the Results table for both benchmarks, hidden by default via a new ``default_hidden`` column flag honored by the column picker.1275bda
-
fix: prefer complete row when deduping config re-runs
CarolinePascalUnverified signatureRank re-runs of the same config by fewest missing (None/NaN) fields first, so an entry with missing items never wins over a complete one; most samples and most recent created_at remain as tie-breakers.2475e35 -
fix: dedupe config re-runs in results table
CarolinePascalUnverified signatureWorkers append a fresh row per benchmark run, so the Hub can hold several rows identical across every config field and differing only in metrics. The Results tab rendered all of them, showing duplicate entries (most visible on the small depth dataset). Collapse re-runs at load time to a single row per config, keeping the one with the most samples and breaking ties by most recent created_at.62848cc
-
refactor: separate depth encoding and quantization params
CarolinePascalUnverified signatureSplit the conflated depth "Encoding" group into distinct Encoding (codec knobs) and Quantization (depth-to-pixel range mapping) groups in the Parameters reference, and regroup the Submit form so lossless (encoding) is separate from use_log + depth range (quantization).292651a -
feat: route all submissions to one repo via benchmark_type
CarolinePascalUnverified signaturePoint the depth benchmark at the shared video-benchmark-submissions repo, always stamp submissions with benchmark_type (rgb/depth), and filter each Submit tab's queue (and its cache key) by benchmark so the two benchmarks coexist in a single submissions dataset.638b995 -
refactor: rename Benchmark dataclass to BenchmarkConfig
CarolinePascalUnverified signaturef78a1a9 -
fix: restrict depth params to implemented set; responsive filters layout
CarolinePascalUnverified signatureDepth codecs -> hevc, libaom-av1; pix_fmt -> gray12le; backend -> pyav only. Filters row now wraps cleanly with per-group min-width and proper row gap instead of squishing/clipping chips and truncating labels on narrow widths.112044f
-
feat: finalize depth submission contract (benchmark_type, *_values booleans, optional ranges)
CarolinePascalUnverified signatureb9c23c9 -
fix: count depth submit axes in cap; surface lossless/use_log in depth leaderboard; doc + fallback cleanups
CarolinePascalUnverified signaturecfd9e58 -
refactor: drop dead _column helper and unused schema import in compute
CarolinePascalUnverified signatureede42e2 -
refactor: parameterize compute.leaderboard and compare_scatter per benchmark
CarolinePascalUnverified signature3e3bac5 -
feat: add DEPTH benchmark config and BENCHMARKS list
CarolinePascalUnverified signaturedc28ce2 -
refactor: add Benchmark config and RGB instance in schema
CarolinePascalUnverified signature420ec0b
-
feat(evaluator): reflecting changes from the evaluator space
CarolinePascalUnverified signature685f1a2
-
chore(leaderboard): polishing leaderboard
CarolinePascalUnverified signature0c60af9
-
feat(gradio): update to gradio app
CarolinePascalUnverified signature1fb78ac