svt_av1_lanczos submission (score: 2.20)

[EthanYangTW]

submission name:

svt_av1_lanczos

upload zipped archive.zip

https://github.com/EthanYangTW/comma_video_compression_challenge/releases/download/v0.2/archive_220.zip

report.txt

=== Evaluation results over 600 samples ===
  Average PoseNet Distortion: 0.08446115
  Average SegNet Distortion: 0.00708126
  Submission file size: 860,695 bytes
  Original uncompressed size: 37,545,489 bytes
  Compression Rate: 0.02292406
  Final score: 100*segnet_dist + √(10*posenet_dist) + 25*rate = 2.20

does your submission require gpu for evaluation (inflation)?

no

did you include the compression script? and want it to be merged?

yes

additional comments

SVT-AV1 with 45% lanczos downscale, CRF 33, preset 0, GOP 180, film-grain=22 (denoise+synth).
Film grain synthesis strips sensor noise before encoding for better compression, decoder re-adds synthetic grain.

[github-actions]

Thanks for the submission @EthanYangTW! 🤏

A maintainer will review your PR shortly.

To run the evaluation, a maintainer will trigger the eval workflow with your PR number.

[Copilot]

Pull request overview

Adds the “winner” submission implementation to the repository, providing an inflation pipeline (decode + resize back to original resolution) and an optional compression script matching the reported leaderboard score.

Changes:

• Added inflate.sh + inflate.py to decode the submission’s .mkv files and upscale frames back to camera_size, writing .raw RGB dumps.
• Added compress.sh that builds archive.zip by encoding videos with SVT-AV1 after Lanczos downscaling.
• Added the submission package initializer (__init__.py) for module execution via python -m submissions.winner.inflate.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
submissions/winner/inflate.sh	Runs the submission’s Python inflator per file listed by the evaluator.
submissions/winner/inflate.py	Decodes video frames with PyAV and resizes to camera_size before writing raw RGB bytes.
submissions/winner/compress.sh	Optional script to generate archive.zip using ffmpeg + SVT-AV1 with downscaling.
submissions/winner/init.py	Makes submissions.winner a Python package for -m execution.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

compress.sh hard-codes the input videos directory and video list file, and doesn’t accept the --in-dir / --video-names-file (and optionally --jobs) CLI flags that other submissions’ compress.sh scripts support (e.g. submissions/baseline_fast/compress.sh). This makes the compression script harder to reuse with other datasets like test_videos.zip and inconsistent with the existing submission script interface.

Consider adding the same argument parsing/usage pattern as baseline_fast so callers can override input dir, names file, and parallelism without editing the script.

Suggested change

	rm -rf "$ARCHIVE_DIR"
	mkdir -p "$ARCHIVE_DIR"
	while IFS= read -r rel; do
	[[ -z "$rel" ]] && continue
	IN="${IN_DIR}/${rel}"
	BASE="${rel%.*}"
	OUT="${ARCHIVE_DIR}/${BASE}.mkv"
	echo "→ ${IN} → ${OUT}"
	ffmpeg -nostdin -y -hide_banner -loglevel warning \
	-r 20 -fflags +genpts -i "$IN" \
	-vf "scale=trunc(iw*0.45/2)*2:trunc(ih*0.45/2)*2:flags=lanczos" \
	-c:v libsvtav1 -preset 0 -crf 32 \
	-g 180 \
	-r 20 "$OUT"
	done < "$VIDEO_NAMES_FILE"
	JOBS=1
	usage() {
	cat <<EOF
	Usage: $(basename "$0") [--in-dir DIR] [--video-names-file FILE] [--jobs N]
	Options:
	--in-dir DIR Input videos directory (default: ${PD}/videos)
	--video-names-file FILE File containing relative video paths (default: ${PD}/public_test_video_names.txt)
	--jobs N Number of parallel ffmpeg jobs to run (default: 1)
	-h, --help Show this help message
	EOF
	}
	while [[ $# -gt 0 ]]; do
	case "$1" in
	--in-dir)
	[[ $# -ge 2 ]] || { echo "Missing value for $1" >&2; usage; exit 1; }
	IN_DIR="$2"
	shift 2
	;;
	--video-names-file)
	[[ $# -ge 2 ]] || { echo "Missing value for $1" >&2; usage; exit 1; }
	VIDEO_NAMES_FILE="$2"
	shift 2
	;;
	--jobs)
	[[ $# -ge 2 ]] || { echo "Missing value for $1" >&2; usage; exit 1; }
	JOBS="$2"
	shift 2
	;;
	-h|--help)
	usage
	exit 0
	;;
	*)
	echo "Unknown argument: $1" >&2
	usage
	exit 1
	;;
	esac
	done
	[[ "$JOBS" =~ ^[1-9][0-9]*$ ]] || { echo "--jobs must be a positive integer" >&2; exit 1; }
	rm -rf "$ARCHIVE_DIR"
	mkdir -p "$ARCHIVE_DIR"
	compress_one() {
	local rel="$1"
	[[ -z "$rel" ]] && return 0
	local in="${IN_DIR}/${rel}"
	local base="${rel%.*}"
	local out="${ARCHIVE_DIR}/${base}.mkv"
	mkdir -p "$(dirname "$out")"
	echo "→ ${in} → ${out}"
	ffmpeg -nostdin -y -hide_banner -loglevel warning \
	-r 20 -fflags +genpts -i "$in" \
	-vf "scale=trunc(iw*0.45/2)*2:trunc(ih*0.45/2)*2:flags=lanczos" \
	-c:v libsvtav1 -preset 0 -crf 32 \
	-g 180 \
	-r 20 "$out"
	}
	export IN_DIR ARCHIVE_DIR
	export -f compress_one
	if [[ "$JOBS" -eq 1 ]]; then
	while IFS= read -r rel; do
	compress_one "$rel"
	done < "$VIDEO_NAMES_FILE"
	else
	xargs -d '\n' -P "$JOBS" -I {} bash -lc 'compress_one "$1"' _ "{}" < "$VIDEO_NAMES_FILE"
	fi

[ylevental]

Great job! I think anyone who wants to beat your score will have to put in substantial effort for a slight improvement.

[YassineYousfi]

@EthanYangTW can you please rename to an informative name like svt_av1_lanczos ?

[github-actions]

Eval Results: winner

=== Evaluation config ===
  batch_size: 16
  device: cpu
  num_threads: 2
  prefetch_queue_depth: 4
  report: submissions/winner/report.txt
  seed: 1234
  submission_dir: submissions/winner
  uncompressed_dir: /home/runner/work/comma_video_compression_challenge/comma_video_compression_challenge/videos
  video_names_file: /home/runner/work/comma_video_compression_challenge/comma_video_compression_challenge/public_test_video_names.txt
=== Evaluation results over 600 samples ===
  Average PoseNet Distortion: 0.33201236
  Average SegNet Distortion: 0.00522368
  Submission file size: 835,231 bytes
  Original uncompressed size: 37,545,489 bytes
  Compression Rate: 0.02224584
  Final score: 100*segnet_dist + √(10*posenet_dist) + 25*rate = 2.90