AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Generative Adversarial Training for Weakly Supervised Cloud Matting

Zhengxia Zou, Wenyuan Li, Tianyang Shi, Zhenwei Shi, Jieping Ye

ICCV 2019 · Point clouds › Segmentation and detection · approach: Point cloud semantic segmentation · 40 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
add adversarial training (ablation)Ours LSGAN+UNet · GF-1 cloud detectionAP avg: 0.9335 -> 0.9523Improved
add saturation penalty to generator objective (ablation)Ours LSGAN+UNet · GF-1 cloud detectiontraining convergenceImproved
add end-to-end training of G, D and F (ablation)Ours LSGAN+UNet · GF-1 cloud detectionAP avg: 0.9430 -> 0.9523Improved
add background image as generator input (ablation)Ours LSGAN+UNet · GF-1 cloud detectionAP avg: 0.9483 -> 0.9523Improved
use LSGAN instead of vanilla GANOurs +UNet · GF-1 cloud detectionAP avg: 0.9462 -> 0.9523Improved
use LSGAN instead of vanilla GANOurs +UNet · GF-1 cloud removalMAE: 0.0720 -> 0.0753Hurt
replace UNet encoder with VGGOurs LSGAN · GF-1 cloud detectionAP avg: 0.9523 -> 0.9538Improved
replace UNet encoder with Resnet50Ours LSGAN · GF-1 cloud detectionAP avg: 0.9523 -> 0.9526No clear effect
compare weakly supervised method with fully supervised UNetOurs LSGAN+UNet · GF-1 cloud detectionAP avg: 0.9613 -> 0.9523Hurt
use proposed method instead of CycleGAN for cloud removalOurs LSGAN+UNet · GF-1 cloud removalMAE: 0.0904 -> 0.0753Improved
add cloud augmentation to trainingSSD (VGG) · Occluded airplane detectionVOC2012-AP: 78.9% -> 81.7%Improved
add cloud augmentation to trainingRetinaNet (Resnet-50) · Occluded airplane detectionVOC2012-AP: 83.3% -> 87.3%Improved

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