all models trained with mnist sigma_data, instead of fashion-mnist.
- base: default k-diffusion model
- no-t-emb: as base, but no t-embeddings in model
- mse-no-t-emb: as no-t-emb, but predicting unscaled noise
- mse: unscaled noise prediction with t-embeddings
base metrics
step,fid,kid 5000,23.366962432861328,0.0060024261474609375 10000,21.407773971557617,0.004696846008300781 15000,19.820981979370117,0.003306865692138672 20000,20.4482421875,0.0037620067596435547 25000,19.459041595458984,0.0030574798583984375 30000,18.933385848999023,0.0031194686889648438 35000,18.223621368408203,0.002220630645751953 40000,18.64676284790039,0.0026960372924804688 45000,17.681808471679688,0.0016982555389404297 50000,17.32500457763672,0.001678466796875 55000,17.74714469909668,0.0016117095947265625 60000,18.276540756225586,0.002439737319946289
mse-no-t-emb
step,fid,kid 5000,28.580364227294922,0.007686138153076172 10000,25.324932098388672,0.0061130523681640625 15000,23.68691635131836,0.005526542663574219 20000,24.05099105834961,0.005819082260131836 25000,22.60521125793457,0.004955768585205078 30000,22.16605567932129,0.0047609806060791016 35000,21.794536590576172,0.0039484500885009766 40000,22.96178436279297,0.005787849426269531 45000,22.641393661499023,0.004763364791870117 50000,20.735567092895508,0.0038640499114990234 55000,21.417423248291016,0.004515647888183594 60000,22.11293601989746,0.0054743289947509766
no-t-emb
step,fid,kid 5000,53.25414276123047,0.02761554718017578 10000,47.687461853027344,0.023845195770263672 15000,46.045196533203125,0.02205944061279297 20000,44.64243698120117,0.020934104919433594 25000,43.55231857299805,0.020574331283569336 30000,43.493412017822266,0.020569324493408203 35000,42.51478958129883,0.01968073844909668 40000,42.213401794433594,0.01972222328186035 45000,40.9914665222168,0.018793582916259766 50000,42.946231842041016,0.019819974899291992 55000,40.699989318847656,0.018331050872802734 60000,41.737518310546875,0.019069194793701172
- Downloads last month
- 0