AI Paper Atlas

Research map / Video / Video generation

An Empirical Study of Autoregressive Pre-training from Videos

Jathushan Rajasegaran, Ilija Radosavovic, Rahul Ravishankar, Yossi Gandelsman, Christoph Feichtenhofer, Jitendra Malik

ICCV 2025 · Video › Video generation · approach: Autoregressive long video generation · 18 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use attention pooling instead of average pooling for probing (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 53.2 -> 61.1Improved
pretrain at 128x128 then finetune at higher resolution with RoPE (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 61.2 -> 63.2Improved
raise RoPE base to 50,000 during finetuning (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 63.2 -> 64.4Improved
use LLaMA architecture instead of GPT2 (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 48.5 -> 53.2Improved
use transformer LLaMA instead of Mamba (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 40.7 -> 53.2Improved
swap tokenizer between dVAE, VQGAN and patch tokens (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 61.3, 61.1, 61.2, 60.6No clear effect
use dVAE tokens as target with patch embedding input (ablation)Toto-large · ImageNet linear probetop-1 accuracy (%): 60.6 -> 58.5Hurt
use full attention instead of causal attention at fine-tuning (ablation)Toto-base · ImageNet fine-tuningtop-1 accuracy (%): 82.2 -> 82.6Improved
add MLP layers to the cross-attention probe head (ablation)Toto-base · Kinetics-400top-1 accuracy (%): 59.3 -> 61.2Improved
scale model from base to 1bToto · Kinetics-400top-1 accuracy (%): 59.3 -> 74.4Improved
scale model from base to 1bToto · ImageNet-1ktop-1 accuracy (%): 64.7 -> 75.3Improved
Toto-base versus MVP encoder for robot cube pickingToto-base · Franka cube picking real worldsuccess rate (%): 75% -> 63%Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy