AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization’s Impact on VLMs Beyond Accuracy

Aymen Bouguerra, Daniel Alfonso Montoya Vasquez, Alexandra Gomez-Villa, Chokri Mraidha, Fabio Arnez

ICML 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Robust fine-tuning for OOD generalization · 1 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
Quantize with QAT (INT8 LSQ) instead of FP32 inferenceCLIP ViT-B/32 (WIT) · CIFAR-100 zero-shotaccuracy (%): 64.47 -> 66.30±0.39Improved
Quantize to INT8 instead of FP16 for noise robustnessCLIP ViT-B/32 (WIT) TensorRT · CIFAR-100 Gaussian bluraccuracy drop (pp): severe blur: 35.64 -> 30.72Improved
Apply rotation (Rot+LSQ) rather than simple PTQ for quantization (ablation)W8A8 quantized VLMs · CounterAnimalAdded vulnerability (%): 3.0 -> -0.2Improved
Add rotation preprocessing to PTQ (ablation)PTQ · Failure rate (>5% accuracy drop)failure rate (%): 62.0% -> 5.6%Improved
Use QAT instead of PTQ (ablation)quantized VLMs · Failure rate (>5% accuracy drop)failure rate (%): 44.1% -> 25.0%Improved
Tune the logit scale after quantization on proxy data (ablation)quantized LAION CLIP models · avg of ImageNet, CIFAR-10/100, SUN397ECE (%): 3.1% -> 1.1%Improved
Quantize the text encoder as well as the vision encoder (PTQ) (ablation)quantized CLIP · calibration, all methodsrelative ECE change: +98% ECEHurt
Quantize models pretrained on noisy LAION dataCLIP LAION models · calibrationrelative ECE change: +49% ECEHurt
Use CC3M proxy data rather than noisy YFCC for logit tuning (ablation)quantized CLIP · proxy dataset choiceECE retentionImproved
Apply W8A8 quantization10 VLMs · synthetic corruptions (CIFAR-10-C)relative robustness gap change (%): +8.9%Improved
Apply W8A8 quantization10 VLMs · natural covariate shift (ImageNet-A/R/V2/Sketch)relative gap change (%): -0.3Hurt
Reduce bit-width from W8A8 to W4A8 with simple PTQ (ablation)quantized VLMs · CounterAnimalDelta RSG (%): 2.6 -> 12.5Hurt

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