AI Paper Atlas

Research map / Detection and segmentation / CNN features

The Lovász-Softmax Loss: A Tractable Surrogate for the Optimization of the Intersection-Over-Union Measure in Neural Networks

Maxim Berman, Amal Rannen Triki, Matthew B. Blaschko

CVPR 2018 · Detection and segmentation › CNN features · approach: Efficient DETR query design · 39 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
Replace cross-entropy with Lovász hingeDeeplabV2 ResNet-101 fine-tuning · Pascal VOC valimage-mIoU: 77.14 -> 80.5Improved
Replace cross-entropy with Lovász-SoftmaxDeeplab-v2 · Pascal VOC testmIoU (%) multi-scale + CRF: 76.44 -> 79.00Improved
Replace cross-entropy with Lovász-Softmax fine-tuningENet · Cityscapes testClass IoU: 58.29 -> 63.06Improved
Fine-tune with Lovász-Softmax instead of cross-entropyDeeplab-based model · IBSR test, 3-fold meanMean Jaccard: 60.16 -> 61.55Improved
Train Lovász-Softmax for 30K rather than 20K iterations (ablation)Deeplab-v2 with equibatch · Pascal VOC valmIoU (%) single-scale: 76.53 -> 77.41Improved
Add equibatch to Lovász-Softmax training (ablation)Deeplab-v2 · Pascal VOC valmIoU (%) multi-scale + CRF: 77.99 -> 78.49Mixed
Add equibatch to cross-entropy training (ablation)Deeplab-v2 · Pascal VOC valmIoU (%) multi-scale + CRF: 76.53 -> 77.31Improved
Optimize per-image IoU rather than batch/dataset IoU (ablation)Lovász-Softmax · Pascal VOCdataset-mIoUHurt
Average the Lovász-Softmax only over classes present in the batch (ablation)Lovász-Softmax · Pascal VOC dataset-mIoUdataset-mIoUImproved
Add Gaussian CRF post-processing (ablation)Lovász-Softmax Deeplab-v2 · Pascal VOC valmIoU (%): 78.22 -> 79.12Improved
Fine-tune ENet with Lovász-SoftmaxENet · Cityscapes testClass iIoU: 34.36 -> 34.06No clear effect
Fine-tune ENet with Lovász-SoftmaxENet · Cityscapes testCat. iIoU: 63.99 -> 61.05Hurt

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