-
Class imbalance: what SynBalance shows
SynBalance (ICCV Workshops 2025) adds generated images for rare classes. On ImageNet-LT, accuracy on the rarest classes rises from 12.6% to 21.4%.
-
Synthetic humans: from SURREAL to BEDLAM
From SURREAL (6.5M frames) to BEDLAM, Michael Black’s line of work shows when synthetic humans match real data, when they don’t, and what’s missing.
-
ROSE: removing objects in video with 3D data
ROSE removes objects from video along with their shadows and reflections, trained on 16,678 rendered 3D pairs. See the numbers and what they don’t prove.
-
DAViD from Microsoft: vision trained on synthetic only
Microsoft’s DAViD trains depth, normals and masks for people on 300K synthetic images. See the numbers, where it trails Sapiens-2B, and the limits.
-
Geo4D: 4D reconstruction trained only on synthetic
Geo4D (ICCV 2025) reconstructs 4D scenes from monocular video, trained only on synthetic data. See the numbers on real video and where it struggles.
-
Infinigen: procedural generation of 3D worlds
Infinigen (Princeton, CVPR 2023) builds natural 3D scenes from rules alone, with no external assets. What the paper measured, what it didn’t, and the cost.
-
LucidSim: generated images to train a robot
MIT CSAIL’s LucidSim (CoRL 2024) trained a quadruped robot in simulation only, using AI-generated images. See the results, real and simulated.
-
Semantic segmentation with synthetic images (2017)
In 2017, 100 Blender images per class raised PASCAL VOC segmentation from 52.80% to 55.47% mIoU. What that proves and what it leaves open to question.
-
Bias in synthetic data: what a 2024 report says
A 2024 report lists the risks and biases of synthetic data, with no experiments of its own. What it says, where it is weak, and how to audit data origin.
-
GAN-generated vs real data: the MIT study
An MIT study (ICLR 2022) trained only on GAN images: below real data on ImageNet, above it on cars. What the study does and doesn’t tell you.
// archive
paper analysis
ArquivoArchive
CategoriasCategories
Posts recentesRecent Posts
TagsTags
4D reconstruction AI and the brain AI ethics AI infrastructure AI transparency BEDLAM bias ChatGPT class imbalance computer vision dataset augmentation depth fundamentals generative models ghost AI Infinigen introduction long-tail recognition MIT study monocular video object removal opinion paper analysis pre-training procedural generation quality inspection real-data limit representation learning representativeness robotics semantic segmentation sim-to-real SMPL SURREAL synthetic data synthetic humans synthetic images tech giants video
