Articles tagged: "datasets"
-
Introducing Stratum-FFHQ: A Multi-Modal Enriched Face Dataset for the Next Generation of Generative Models
Every researcher training generative models on faces knows the drill: download FFHQ (Karras et al., 2019), write a preprocessing pipeline, run inference with a segmentation model, compute depth maps, extract... -
The Literal Value Bottleneck: Why GNN Autoencoders Fail at Code Generation
We trained GNN autoencoders on 22,000 Ruby ASTs. The models achieved 81% node type accuracy and 99.5% type diversity, yet generated exactly 0% syntactically valid code. Here is why.