Benchmarks — VIDing.AI

Test Environment

Hardware

AMD EPYC 9654 / 96 cores

FFmpeg

7.1 (libx264, libx265, libaom-av1)

Test Corpus

12 clips × 3 resolutions

Measurement

VMAF v0.6.1 / libvmaf

Rate-Distortion Curves

4K UHD — VMAF vs Bitrate

1080p — VMAF vs Bitrate

Encoding Speed (FPS) by Codec & Resolution

BD-Rate Savings vs H.264 Baseline

4K UHD — Balanced Preset

Codec	CRF	Speed	VMAF	PSNR (dB)	SSIM	Size Reduction	Encode Speed
H.264	23	medium	93.2 ±0.4	42.1 ±0.3	0.9891	62%	28 fps
H.265	28	medium	93.4 ±0.3	41.8 ±0.4	0.9884	74%	12 fps
AV1	30	4	93.1 ±0.5	42.3 ±0.3	0.9893	79%	2.4 fps

1080p — Quality Preset

Codec	CRF	Speed	VMAF	PSNR (dB)	SSIM	Size Reduction	Encode Speed
H.264	18	slow	97.4 ±0.2	46.8 ±0.2	0.9962	48%	42 fps
H.265	22	slow	97.6 ±0.2	46.5 ±0.3	0.9958	61%	18 fps
AV1	26	3	97.3 ±0.3	47.1 ±0.2	0.9964	68%	4.1 fps

720p — Speed Preset

Codec	CRF	Speed	VMAF	PSNR (dB)	SSIM	Size Reduction	Encode Speed
H.264	28	ultrafast	88.6 ±0.6	38.4 ±0.4	0.9782	71%	185 fps
H.265	32	ultrafast	89.1 ±0.5	38.1 ±0.5	0.9774	78%	92 fps
AV1	35	8	88.9 ±0.6	38.7 ±0.4	0.9791	83%	14 fps

Methodology: All tests use the same 12-clip corpus covering aerial, interview, sports, animation, screen recording, and cinematic content at each resolution. Each configuration is run 5 times; values show the mean with ±95% confidence intervals. VMAF scores are computed per-frame and averaged. Encode speed measured on a single worker thread. Size reduction is compared against the source bitrate. Results may vary based on content type and source encoding.

Next-Generation Pipeline

Selective Encoding Enhancement Results

VIDing.AI's patented CIP (Content-Intelligent Processing) system analyzes each frame to apply encoding decisions selectively, achieving dramatically better speed-quality tradeoffs.

3.22×

Encoding Speedup

Combined system vs baseline encoder
Quality within 0.1 dB of reference

10×

Entropy Estimation Speed

Multi-scale estimation: 2.1 ms → 0.18 ms/frame
98.3% correlation with full-scale

+0.8 dB

PSNR Improvement

Confidence-driven adaptive quantization
7.1% bitrate reduction at equal quality

88.9%

Battery Life Gain

Mobile playback scenario
Reduced decoder complexity + smaller files

Component Breakdown

Component	Metric	Baseline	CIP Enhanced	Improvement
Multi-Scale Entropy Estimation	Processing time / frame	2.1 ms	0.18 ms	10× faster
	Correlation with full-scale	1.000 (reference)	0.983	98.3% fidelity
Adaptive Quantization	PSNR (dB)	42.1	42.9	+0.8 dB
	Bitrate at equal quality	baseline	-7.1%	7.1% savings
Combined CIP System	Encoding speed	1.0×	3.22×	222% faster
	Quality delta (PSNR)	reference	-0.1 dB	Visually identical
Mobile Decode Efficiency	Battery consumption (playback)	baseline	11.1%	88.9% reduction

Test conditions: CIP measurements performed on the same 12-clip corpus at 4K UHD resolution using H.264 baseline encoder. Entropy estimation benchmarked on per-frame basis (n=10,000 frames). Adaptive quantization evaluated at CRF 23 medium preset. Battery measurements from Qualcomm Snapdragon 8 Gen 3 reference platform, 60-minute continuous playback loop. PSNR delta of ±0.1 dB is below perceptual threshold (ITU-R BT.500-14).

Encoder Benchmarks