A list of public training logs from neural network image generation models, since I think they're interesting.
- Publicly accessible link
- Losses plotted every so often
- Samples generated every so often
- Nontrivial dataset (i.e. not MNIST - 64x64 output RGB or better)
- Nontrivial runtime (i.e. the final images actually look like something)
- Trained from scratch (i.e. step 0 samples must be outputs from randomly-initialized network)
- wandb
- Open Muse (various masked latent generator training runs) https://wandb.ai/psuraj/muse
- Latent Diffusion training on COCO (CompVis/latent-diffusion#132 (comment)) https://wandb.ai/yangxingyi/uncategorized/runs/2022-09-10T13-25-40_txt2img-coco-kl-8-128dim_2e-7_repeat10new?workspace=user-yangxingyi
- DALL E Mega training journal (a latent autoregressive model) https://wandb.ai/dalle-mini/dalle-mini/reports/DALL-E-Mega-Training-Journal--VmlldzoxODMxMDI2
- theseriousadult's diffusion txt2img runs https://wandb.ai/theseriousadult/diffusion-txt2img/runs/3svj2fed?workspace=user-
- JD-P's cloob-conditioned latent diffusion runs (https://github.com/JD-P/cloob-latent-diffusion/ / https://wandb.ai/jdp/jdp-latent-diffusion?workspace=user-)
- YFCC flickr photos https://wandb.ai/jdp/jdp-latent-diffusion/runs/1dv7xxrg?workspace=user-jdp https://wandb.ai/jdp/jdp-latent-diffusion/runs/258cmlpw?workspace=user-jdp
- danbooru anime images https://wandb.ai/jdp/jdp-latent-diffusion/runs/2nr62ymp?workspace=user-
- wikiart https://wandb.ai/jdp/jdp-latent-diffusion/runs/35pqdn83?workspace=user-
64x64 laion400m imagen-pytorch conditional diffusion report https://wandb.ai/zenforic/imagen-pt-zen/reports/Partial-Laion-400M-on-Imagen--VmlldzoyMTgxNTMw(lots of good notes here - but I can't find samples)- 64x64 laion400m imagen-pytorch conditional diffusion https://wandb.ai/camlaedtke/imagen/runs/3oykwi7g?workspace=user- (this one has samples - also see discussion in lucidrains/imagen-pytorch#89)
- 64x64 GLIDE on a bunch of images https://wandb.ai/afiaka87/laion_wikart_cc12m_coco_vg_openimages_faraday/runs/2lcdo9ol?workspace=user-
- conditional denoising diffusion GAN https://wandb.ai/johnowhitaker/cclddg_report/reports/New-Techniques-for-Generating-Images-With-Text--VmlldzoxOTc1NDU3
- conditional cifar 10 DDPM https://wandb.ai/capecape/train_sd/reports/How-To-Train-a-Conditional-Diffusion-Model-From-Scratch--VmlldzoyNzIzNTQ1
- huggingface (all unconditional ddpm)
- 64x64 pokemon https://huggingface.co/anton-l/ddpm-ema-pokemon-64/tensorboard?params=images#frame
- 64x64 oxford flowers https://huggingface.co/anton-l/ddpm-ema-flowers-64/tensorboard?params=images#frame
- 64x64 pets https://huggingface.co/pcuenq/ddpm-ema-pets-64/tensorboard?params=images#frame
- 64x64 pets (no timestep conditioning - glitchy) https://huggingface.co/pcuenq/ddpm-ema-pets-64-no-tcond/tensorboard?params=images#frame
- cifar10 https://huggingface.co/pcuenq/ddpm-ema-cifar/tensorboard?params=images#frame
- 64x64 faces https://huggingface.co/pcuenq/ddpm-ema-faces-64-50/tensorboard?params=images#frame
- more 64x64 faces https://huggingface.co/sreerajr000/ddpm-ema-fairface-64/tensorboard?params=images#frame
- 128x128 anime faces https://huggingface.co/mrm8488/ddpm-ema-anime-128/tensorboard?params=images#frame (also https://huggingface.co/mrm8488/ddpm-ema-anime-v2-128/tensorboard?params=images#frame)
- 256x256 anime faces https://huggingface.co/mrm8488/ddpm-ema-anime-256/tensorboard?params=images#frame
- 128x128 butterflies https://huggingface.co/mrm8488/ddpm-ema-butterflies-128/tensorboard?params=images#frame
- kaggle
- 115k laion text2image latent diffusion training in keras (using SD's VAE) https://www.kaggle.com/code/apapiu/train-latent-diffusion-in-keras-from-scratch