HolosView R&D Dashboard

Phase

1 — Channel Packing

Byte-exact decoding verified. 4 packing strategies benchmarked.

Baseline Size

483.5 KiB

Garden XXS · 32K splats · 9 WebP planes · 15.5 B/gaussian

Round-trip

✓ Byte-Exact

All channels pass except quats (norm issue) and merged alpha planes

Test Scene

Mip-NeRF 360 Garden

XXS variant · SOG-XT v3 · Not an ecommerce product

📊 Packing Benchmark Results

Strategy	Planes	Encoded	KiB	B/gauss	Ratio	RT
Baseline SOG-XT	9	495,112	483.5	15.5	2.57x	partial
Merged Opacity+Scales	8	494,724	483.1	15.5	2.30x	partial
Geometry Packed	7	495,734	484.1	15.5	2.02x	partial
Minimal 6-plane	7	495,734	484.1	15.5	2.02x	partial

📏 Per-Plane Entropy (Baseline)

Plane	Raw Bytes	Encoded	Ratio	Entropy	RT
active_mask	135,424	156	868.1x	0.23 b/sym	✓
means_bytes_0 (lo)	135,424	86,942	1.56x	5.88 b/sym	✓
means_bytes_1 (hi)	135,424	45,138	3.00x	5.24 b/sym	✓
opacities	135,424	24,612	5.50x	4.55 b/sym	✓
scales	135,424	78,542	1.72x	4.94 b/sym	✓
quaternions	135,424	68,460	1.98x	4.82 b/sym	⚠
f_dc (SH base)	135,424	70,994	1.91x	5.00 b/sym	✓
sh_centroids	188,160	76,510	2.46x	4.10 b/sym	✓
sh_labels	135,424	43,758	3.09x	2.86 b/sym	✓

🧬 Hypotheses Status

TESTINGH1: Adaptive RGBA channel allocation beats per-attribute planes

PENDINGH2: Block-local spatial ordering beats globally sorted layouts

PENDINGH3: Base+residual prediction improves size/decoding

PENDINGH4: Shared texture/codebook tiles reduce total bits

PENDINGH5: Shader-side quantised decoding improves startup/runtime

PENDINGH6: Product-view-aware pruning removes splats while preserving quality

PENDINGH7: Progressive model reaches interactivity sooner

PENDINGH8: Product-specific encoder/player wins on total performance

🖥️ Environment

CPU

AMD Ryzen AI Max+ 395

16 cores / 32 threads · 122 GB RAM

GPU

Radeon 8060S (RADV)

Mesa 26.0.8 · Vulkan 1.4.335 · No CUDA

Stack

Python 3.11 · Node 22 · Wrangler

NumPy · Pillow · SciPy · Cloudflare Pages/Workers/R2/D1

📂 Key Findings

🔴 Channel merging does NOT help: Combining opacity+scales into one RGBA plane saves only 0.4 KiB (0.08%). WebP lossless compresses correlated data better when kept separate.
🟢 mask plane is massively compressible: 868x ratio — nearly all texels are active. Can be represented as a simple count or RLE.
🟢 SH labels compress well: 3.09x at only 2.86 b/sym entropy — only 56 unique centroids for 32K gaussians.
🟡 Position (means) hi-byte compresses 2x better than lo-byte: 3.00x vs 1.56x — spatial coherence in high bytes.
🟡 Quaternion norms not 1.0: Mean norm = 0.6086 after dequantization. Possible encoding/normalization artifact. Needs investigation.
🔴 Merged planes increase entropy: opacity+scales merged = 6.46 b/sym (vs 4.55 + 4.94 separate). Mixing uncorrelated data hurts WebP compression.
⚠️ Alpha passthrough issue: Quats and merged planes with alpha data show round-trip failures in some channels. WebGL/WebGPU alpha handling needs investigation.

⏭️ Next Steps (Phase 2)

1. Investigate quaternion norm issue and alpha passthrough in WebP round-trips
2. Implement spatial ordering (Morton, Hilbert) on actual Gaussian positions
3. Test block-local coordinate encoding (cell origin + local offsets)
4. Decode HolosView PLY files and test on real product reconstructions
5. Build and deploy viewer with side-by-side comparison

🔬 HolosView R&D Dashboard