the archive

#loss functions

8 results

cs.CV2019

IoU Loss for 2D/3D Object Detection

Dingfu Zhou, Jin Fang, Xibin Song +4

The paper introduces a unified IoU‑based loss function that can be applied to both axis‑aligned and rotated 2D/3D bounding boxes, reducing the gap between training and testing perf…

#object detection#intersection-over-union#rotated bounding boxes#3d detection

eess.IV2019

Distance Map Loss Penalty Term for Semantic Segmentation

Francesco Caliva, Claudia Iriondo, Alejandro Morales Martinez +2

The paper proposes a distance‑map based loss penalty term to guide convolutional networks toward better boundary accuracy in semantic segmentation, demonstrating improved bone shap…

#semantic segmentation#loss functions#distance maps#bone segmentation

cs.LG2019

Improving Adversarial Robustness via Guided Complement Entropy

Hao-Yun Chen, Jhao-Hong Liang, Shih-Chieh Chang +4

The paper introduces Guided Complement Entropy, a new loss function that improves adversarial robustness without extra training steps, while also boosting standard accuracy, and ca…

#adversarial robustness#loss functions#deep learning#computer vision

cs.LG2019

Toward a Characterization of Loss Functions for Distribution Learning

Nika Haghtalab, Cameron Musco, Bo Waggoner

The paper investigates loss functions for learning probability distributions over large discrete domains, proposes desirable criteria for such losses, and shows that while no loss…

#loss functions#distribution learning#density estimation#calibrated distributions

cs.CV2019

Enforcing geometric constraints of virtual normal for depth prediction

Wei Yin, Yifan Liu, Chunhua Shen +1

The paper proposes a loss that enforces virtual normal constraints—directions defined by three random points in the reconstructed 3D space—to improve monocular depth prediction, re…

#monocular depth estimation#geometric constraints#virtual normals#3d reconstruction

cs.CV2019

Competing Ratio Loss for Discriminative Multi-class Image Classification

Ke Zhang, Xinsheng Wang, Yurong Guo +3

The paper proposes a new loss function called Competing Ratio Loss (CRL) for training deep convolutional neural networks on multi‑class image classification, aiming to better separ…

#image classification#loss functions#deep learning#convolutional neural networks

cs.CV2019

Content and Colour Distillation for Learning Image Translations with the Spatial Profile Loss

M. Saquib Sarfraz, Constantin Seibold, Haroon Khalid +1

The paper introduces a Spatial Profile Loss that directly compares source and target images to distill shape/content and colour/style, allowing GAN-based image translation without…

#image translation#generative adversarial networks#loss functions#content distillation

eess.IV2019

Blind Deblurring Using GANs

Manoj Kumar Lenka, Anubha Pandey, Anurag Mittal

The paper explores various GAN architectures for blind image deblurring, introducing attention modules, residual connections, and combined loss functions to improve global image pe…

#blind deblurring#generative adversarial networks#attention mechanisms#residual connections