text stringlengths 861 5.87k |
|---|
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Patrick Esser*Sumith Kulal Andreas Blattmann Rahim Entezari Jonas M ¨uller Harry Saini Yam Levi Dominik Lorenz Axel Sauer Frederic Boesel Dustin Podell Tim Dockhorn Zion English Kyle Lacey Alex Goodwin Yannik Marek Robin Rombach* Stability AI Figur... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis 1. Introduction Diffusion models create data from noise (Song et al., 2020). They are trained to invert forward paths of data towards random noise and, thus, in conjunction with approximation and generalization properties of neural networks, can be... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Fora0= 1, b0= 0, a1= 0andb1= 1, the marginals, pt(zt) =Eϵ∼N(0,I)pt(zt|ϵ), (3) are consistent with the data and noise distribution. To express the relationship between zt, x0andϵ, we intro-duceψtandutas ψt(·|ϵ) :x07atx0+btϵ (4) ut(z|ϵ):=ψ′ t(ψ-1 t(z... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (LDM-)Linear LDM (Rombach et al., 2022) uses a mod-ification of the DDPM schedule (Ho et al., 2020). Both are variance preserving schedules, i. e. bt=p 1-a2 t, and de-fineatfor discrete timesteps t= 0,..., T-1in terms of diffusion coefficients βtas... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Caption CLIP-L/14 CLIP-G/14 T5 XXLPooled Linear c MLP MLP Sinusoidal Encoding Timestep+ y Noised Latent Patching Linear +Positional Embedding x MM-Di T-Block 1 MM-Di T-Block 2... MM-Di T-Block d Modulation Linear Unpatching Output77 + 77 tokens 409... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis rank averaged over variant all 5 steps 50 steps rf/lognorm(0. 00, 1. 00) 1. 54 1. 25 1. 50 rf/lognorm(1. 00, 0. 60) 2. 08 3. 50 2. 00 rf/lognorm(0. 50, 0. 60) 2. 71 8. 50 1. 00 rf/mode(1. 29) 2. 75 3. 25 3. 00 rf/lognorm(0. 50, 1. 00) 2. 83 1. 50 2... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis 10 20 30 40 506080100120140 edm(-1. 20, 1. 20) eps/linear rf/lognorm(0. 00, 1. 00) rf v/cos v/linear number of sampling steps FID Figure 3. Rectified flows are sample efficient. Rectified Flows perform better then other formulations when sampling f... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis a space elevator, cinematic scifi art A cheeseburger with juicy beef patties and melted cheese sits on top of a toilet that looks like a throne and stands in the middle of the royal chamber. a hole in the floor of my bathroom with small gremlins li... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 4. Training dynamics of model architectures. Compara-tive analysis of Di T,Cross Di T,UVi T, and MM-Di T on CC12M, focusing on validation loss, CLIP score, and FID. Our proposed MM-Di T performs favorably across all metrics. shared set of we... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 6. Timestep shifting at higher resolutions. Top right: Hu-man quality preference rating when applying the shifting based on Equation (23). Bottom row: A5122model trained and sam-pled withp m/n = 1. 0(top) andp m/n = 3. 0(bottom ). See Sectio... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis After the shifted training at resolution 1024×1024, we align the model using Direct Preference Optimization (DPO) as described in Appendix C. 5. 3. 3. R ESULTS In Figure 8, we examine the effect of training our MM-Di T at scale. For images, we cond... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 8. Quantitative effects of scaling. We analyze the impact of model size on performance, maintaining consistent training hyperparameters throughout. An exception is depth=38, where learning rate adjustments at 3×105steps were necessary to pre... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Broader Impact This paper presents work whose goal is to advance the field of machine learning in general and image synthesis in par-ticular. There are many potential societal consequences of our work, none of which we feel must be specifically hig... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Elsayed, G. F., Mahendran, A., Yu, F., Oliver, A., Huot, F., Bastings, J., Collier, M. P., Gritsenko, A., Birodkar, V., Vasconcelos, C., Tay, Y., Mensink, T., Kolesnikov, A., Paveti ´c, F., Tran, D., Kipf, T., Lu ˇci´c, M., Zhai, X., Keysers, D., H... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis 10. 1007/978-3-319-10602-1 48. URL http://dx. d oi. org/10. 1007/978-3-319-10602-1_48. Lipman, Y., Chen, R. T. Q., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modeling. In The Eleventh International Conference on Learning Repr... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Saharia, C., Chan, W., Chang, H., Lee, C., Ho, J., Salimans, T., Fleet, D., and Norouzi, M. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 Conference Proceedings, pp. 1-10, 2022a. Saharia, C., Chan, W., Saxena, S., Li, L., Whang, J.... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Supplementary A. Background Diffusion Models (Sohl-Dickstein et al., 2015; Song et al., 2020; Ho et al., 2020) generate data by approximating the reverse ODE to a stochastic forward process which transforms data to noise. They have become the stand... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Detailed pen and ink drawing of a happy pig butcher selling meat in its shop. a massive alien space ship that is shaped like a pretzel. A kangaroo holding a beer, wearing ski goggles and passionately singing silly songs. An entire universe inside a... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis detailed pen and ink drawing of a massive complex alien space ship above a farm in the middle of nowhere. photo of a bear wearing a suit and tophat in a river in the middle of a forest holding a sign that says ”I cant bear it”. tilt shift aerial ph... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis B. On Flow Matching B. 1. Details on Simulation-Free Training of Flows Following (Lipman et al., 2023), to see that ut(z)generates pt, we note that the continuity equation provides a necessary and sufficient condition (Villani, 2008): d dtpt(x) +∇ ... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 10. FID scores after training flow models with different sizes (parameterized via their depth) on the latent space of different autoencoders (4 latent channels, 8 channels and 16 channels) as discussed in Section 5. 2. 1. As expected, the fl... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 11. The mode (left) and logit-normal (right) distributions that we explore for biasing the sampling of training timesteps. “A raccoon wearing formal clothes, wearing a tophat and holding a cane. The raccoon is holding a garbage bag. Oil pain... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis C. Direct Preference Optimization “a peaceful lakeside landscape with migrating herd of sauropods”“a book with the words 'Don't Panic¡, written on it”2B base 2B w/ DPO 8b base 8b w/ DPO Figure 13. Comparison between base models and DPO-finetuned mo... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Prompt Quality0102030405060Human Preference [ % ]depth=24 (2B) base w/ DPO Prompt Qualitydepth=38 (8B) base w/ DPO Figure 14. Human preference evaluation between base models and DPO-finetuned models. Human evaluators prefer DPO-finetuned models for... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Input Output 1 Output 2 Write ”go small go home” instead GO BIG OR GO UNET is written on the blackboard change the word to UNOT make the sign say MMDIT rules Figure 15. Zero Shot Text manipulation and insertion with the 2B Edit model Details on Ded... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis E. 3. Assessing the Efficacy of our Deduplication Efforts Carlini et al. (2023) devise a two-stage data extraction attack that generates images using standard approaches, and flags those that exceed certain membership inference scoring criteria. Ca... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (a) Final result of SSCD deduplication over the entire dataset (b) Result of SSCD deduplication with various thresholds over 1000 random clusters Figure 16. Results of deduplicating our training datasets for various filtering thresholds. Algorithm ... |
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis Figure 17. SSCD-based deduplication prevents memorization. To assess how well our SSCD-based deduplication works, we extract memorized samples from small, specifically for this purpose trained models and compare them before and after deduplication.... |
README.md exists but content is empty.
- Downloads last month
- 12