AI Paper Atlas

Research map / Architectures and efficiency / Quantization

NVTC: Nonlinear Vector Transform Coding

Runsen Feng, Zongyu Guo, Weiping Li, Zhibo Chen

CVPR 2023 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 19 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use entropy-constrained VQ rather than nonlinear transform coding with scalar quantizationToy-source quantization · Isotropic Gaussian 2d, 4d, 8d, and 16dBD-PSNR (dB): -0.15, -0.31, -0.47, and -0.71 BD-PSNR for NTC over ECVQImproved
Use entropy-constrained VQ rather than nonlinear transform coding with scalar quantizationToy-source quantization · IG, Banana, and BoomerangBD-PSNR (dB): -0.15, -0.21, and -0.58 BD-PSNR for NTC over ECVQImproved
Replace VT with convolutional resblocks (ablation)NVTC · CLIC 2021 validsetBD-rate: 0.0% -> 7.3%Hurt
Replace ECVQ with unconstrained VQ (ablation)NVTC · CLIC 2021 validsetBD-rate: 0.0% -> 23.9%Hurt
Use random initialization instead of progressive initialization (ablation)NVTC · CLIC 2021 validsetBD-rate: 0.0% -> 17.1%Hurt
Reduce quantization-layer groups to L=3+3+2 (ablation)NVTC · KodakBD-rate: 0.0% -> 7.7%Hurt
Remove the first quantization-layer group, using L=0+6+4 (ablation)NVTC · KodakBD-rate: 0.0% -> 6.8%Hurt
Use only the final quantization-layer group, L=0+0+4 (ablation)NVTC · KodakBD-rate: 0.0% -> 27.1%Hurt
Replace Depth-wise Block FC with Block FC (ablation)NVTC · KodakBD-rate: 0.0% -> 29.4%Hurt
Replace CEM with an unconditional entropy model in entropy coding (ablation)NVTC · KodakBD-rate: 0.0% -> 39.6%Hurt
Use NVTC rather than Zhu-CVPR2022Learned image compression · CLIC2021 validsetBD-rate: -10.9%Improved

Later papers that compared with this method

1 later accepted paper reports NVTC as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy