datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
exp03-l23-hardening
Exp 03 / 03b — Klüver L2/3 hardening study (SDXL + SD 3.5)
Pre-registered study from the Operating System Hypothesis project.
Sweep classifier-free guidance across two architectures and score every output
blind, on two independent rubrics, for how far object structure has come apart.
The prediction was written down and committed before the run. The commit
dates in the GitHub repo are the proof.
What is here that is not on GitHub
The 860 generated PNGs. Every text… See the full description on the dataset page: https://huggingface.co/datasets/youssefhassan13/exp03-l23-hardening.bongard-rwr-plus-l2
Bongard RWR+/L2
This is a variant of the dataset featuring two images per side. The original dataset, which includes six images per side, can be found here.
Bongard Problems (BPs) provide a challenging testbed for abstract visual reasoning (AVR), requiring models to identify visual concepts from just a few examples and describe them in natural language.
Early BP benchmarks featured synthetic black-and-white drawings, which might not fully capture the complexity of real-world scenes.… See the full description on the dataset page: https://huggingface.co/datasets/spawlonka/bongard-rwr-plus-l2.L2P-dataset
L2P: Unlocking Latent Potential for Pixel Generation
An efficient transfer paradigm enabling high-quality, end-to-end pixel-space diffusion with minimal computational overhead and data requirements.
Pixel diffusion models have recently regained attention for visual generation. However, training advanced pixel-space models from scratch demands prohibitive computational and data resources. To address this, we propose the Latent-to-Pixel (L2P) transfer… See the full description on the dataset page: https://huggingface.co/datasets/zhen-nan/L2P-dataset.CLIP-ViT-L-14-336-L20-features
OpenAI/CLIP-ViT-L/14@336 Layer 20 features, CLIP+BLIP labels
Feature activation max visualization of the 4096 Features @ L20
CLIP+BLIP labels (may or may not describe what a neuron truly encodes!)
⚠️ May contain sensitive images, albeit abstract. Use responsibly!
Examples:
wiki10-31k-l2rboot-llama8b-ckptdreamsim_crop_cosine-gpt4_diverse_prompts_NG_NT_NKE5_NKCO50_L2B5dreamsim_crop_cosine-gpt4_DE_diverse_prompts_NG_NT_NKE5_NKCO50_L2B5metamdp-robosuite-franka-moving_ball-l2-structured-train-state16-h50-v1l2p-datasetLatent-to-pixel Qwen-Image-2512 synthetic run (raw)
20k prompts x 2 seeds = 40k samples
prompts generated using Qwen3.6
shirakami_fubuki_illust
Shirakami Fubuki Illustrations Dataset
A curated dataset of high-quality illustrations featuring Shirakami Fubuki (白上フブキ), a popular virtual YouTuber from Hololive.
Dataset Description
This dataset contains original artwork and fan art of Shirakami Fubuki, collected from Danbooru, a popular anime-style image board. All images are tagged with comprehensive metadata including artist information, character tags, and copyright details.
Key Features
High-Quality… See the full description on the dataset page: https://huggingface.co/datasets/l2533584225/shirakami_fubuki_illust.l2p-clean
L2P-Clean — curated (prompt, image) pairs for Latent-to-Pixel transfer on Qwen-Image-2512
Aesthetic-curated synthetic dataset for the L2P distillation recipe
(arXiv:2605.12013): the source diffusion model's
own outputs, cleaned "to the bones" so the pixel decoder fits an already-smooth,
high-quality teacher manifold.
Provenance
Merged and de-conflicted from three independent synthetic runs (each its own prompt
corpus — prompt ids are namespaced with a raw_ /… See the full description on the dataset page: https://huggingface.co/datasets/shauray/l2p-clean.deadly_corridor_fixed_l2_fs6netdsl-l2-synthetic-sem
NetDSL-L2 Synthetic SEM Circuit Patterns (v3)
Synthetic image–DSL training corpus for the EUSIPCO 2026 paper
"Bridging the Sim-to-Real Gap in Semiconductor Visual Program Synthesis via Input Binarization"
(Ohtsubo, Dohi, Yawata, Takeshita, Sasaki — Hitachi, Ltd. R&D Group).
This is the only data the model is trained on. No real SEM image is used during training;
the sim-to-real transfer is handled entirely at inference time by input binarization.
📄 Paper & code:… See the full description on the dataset page: https://huggingface.co/datasets/utsubo12/netdsl-l2-synthetic-sem.MVSA-Ml2p-part0demon_attack_fixed_l2_fs1demon_attack_fixed_l2_fs2HCM-AIC-keyframes-L29_abongard-rwr-plus-l2
Bongard RWR+/L2
This is a variant of the dataset featuring two images per side. The original dataset, which includes six images per side, can be found here.
Bongard Problems (BPs) provide a challenging testbed for abstract visual reasoning (AVR), requiring models to identify visual concepts from just a few examples and describe them in natural language.
Early BP benchmarks featured synthetic black-and-white drawings, which might not fully capture the complexity of real-world scenes.… See the full description on the dataset page: https://huggingface.co/datasets/bongard-rwr-plus/bongard-rwr-plus-l2.deadly_corridor_fixed_l2_fs4_10k_epsdreamsim_crop_cosine-targets_60words_DE_NG_NT_NKCO50_L2B5demon_attack_fixed_l2_fs6Complex-Images-L2-DatasetDataset Description:
This dataset is a large-scale collection of Complex Images L2, containing 15,023 images, designed to support the development and training of advanced computer vision, image recognition, visual reasoning, and multimodal AI systems.
Additionally, this dataset can be used in pipelines for Supervised Fine-Tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF) workflows, improving model performance in image understanding, scene interpretation, object recognition… See the full description on the dataset page: https://huggingface.co/datasets/InfoBayAI/Complex-Images-L2-Dataset.Doc-Mix-L224MASADdreamsim_crop_cosine-targets_60words_DE_NG_NT_NKE5_NKCO50_L2B5deadly_corridor_fixed_l2_fs4flappy_fixed_l2_fs4dreamsim_crop_cosine-targets_NG_NT_NKCO50_L2B5dreamsim_crop_cosine-targets_NG_NT_NKE5_NKCO50_L2B5