AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

EVC: Towards Real-Time Neural Image Compression with Mask Decay

Wang Guo-Hua, Jiahao Li, Bin Li, Yan Lu

ICLR 2023 · Architectures and efficiency › RNNs and normalization · approach: Dataset condensation and flow estimation · 23 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
Apply mask decay to the Medium-Medium model (ablation)EVC Medium-Medium · KodakBD-Rate (%) vs VTM: 2&9 -> 0&9Improved
Apply mask decay to the Small-Small model (ablation)EVC Small-Small · KodakBD-Rate (%) vs VTM: 10&0 -> 6&9Improved
Use residual representation learning instead of training separate encoders (ablation)Scalable-EVC with Large decoder · Kodak, S0,1,2,3BD-Rate (%) vs VTM: 1.5 -> 0.6Improved
Add mask decay to residual representation learning (ablation)Scalable-EVC with Large decoder · Kodak, S0,1,2,3BD-Rate (%) vs VTM: 0.6 -> -0.2Improved
Add mask decay to residual representation learning (ablation)Scalable-EVC with Small decoder · Kodak, S0,1,2,3BD-Rate (%) vs VTM: 5.5 -> 4.8Improved
Use the proposed sparsity loss rather than L1 or L2 regularization (ablation)EVC mask decay · KodakBD-Rate (%) vs VTMImproved
Set mask decay rate to 8e-5 rather than the suitable 4e-5 (ablation)EVC mask decay · KodakBD-Rate (%) vs VTMHurt
Use L2 sparsity loss for mask decay (ablation)EVC mask decay · KodakBD-Rate (%) vs VTMHurt
Avoid channels selected by the raw mask-decay procedure (ablation)Avoidable Mask Decay · KodakBD-Rate (%) vs VTMNo clear effect
Compress the encoder rather than the decoder (ablation)EVC with a Large counterpart module · KodakBD-Rate (%) loss: 2.7% vs 5.1%Improved
Use EVC Small rather than STF CNNImage compression at 768×512 on A100 · 768×512encoding latency (ms): 96.4 -> 17.7Improved

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