Visual Understanding and Generative Modeling
(Fall 2025)

Summary

This course is a student-led seminar course, focusing on advanced topics in the machine learning-based computer vision area. Specifically, we cover recent machine perception and representation learning methods in the half of this course (on Thursdays), and recently developed generative models for images and videos, including multimodal models, in the rest half (on Tuesdays). Each day, we will cover three papers, with a student presentation and discussion moderated by the instructor. All students are expected to contribute to this course by presenting 2-3 papers and actively participate in the discussion, including asking questions.

Logistics

  • Organizers:

    Joonseok Lee
    Instructor

    Junho Lee
    TA
  • Day/Time: Tuesdays and Thursdays, 2:00 pm - 3:15 pm
  • Classroom: 942 - 302
  • Prerequisite:
    • A student taking this course must have successfully taken Machine Learning for Visual Understanding course.
    • If not, but have taken another equivalent graduate-level computer vision course, an explicit approval for qualification is needed from the instructor.
    • As this course is a seminar-based one, it is important to ensure all participants are prepared to present recent papers with sufficient prior knowledge.
  • Grading: Attendance and participation 100%

Syllabus

Date
Topic
Link
Presenter
9/2
Tue
Course Introduction
  • About the course
  • Logistics and grading
9/4
Thr
Transformer architectures for visual understanding I
  • Transformers are Graph Neural Networks
  • Energy-Based Transformers are Scalable Learners and Thinkers
  • MambaVision: A Hybrid Mamba-Transformer Vision Backbone (CVPR'25)


  • TBD
  • TBD
  • TBD
9/9
Tue
Diffusion Transformers
  • Reconstruction vs. generation: Taming optimization dilemma in latent diffusion models (CVPR'25)
  • EDiT: Efficient Diffusion Transformers with Linear Compressed Attention (ICCV'25)
  • MambaVision: A Hybrid Mamba-Transformer Vision Backbone (CVPR'25)


  • TBD
  • TBD
  • TBD
9/11
Thr
Transformer architectures for visual understanding II
  • Differential Transformer (ICLR'25)
  • Spiking Vision Transformer with Saccadic Attention (ICLR'25)
  • On the rankability of visual embeddings


  • TBD
  • TBD
  • TBD
9/16
Tue
No Class (IFSCC)
9/18
Thr
No Class (IFSCC)
9/23
Tue
Diffusion models for image generation I
  • Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models (ICLR'25)
  • SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer (ICML'25)
  • Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR'25)


  • TBD
  • TBD
  • TBD
9/25
Thr
Training vision transformer models
  • LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributions (CVPR'25)
  • Why Gradients Rapidly Increase Near the End of Training (arXiv'25)
  • Progressive distillation induces an implicit curriculum (ICLR'25)


  • TBD
  • TBD
  • TBD
9/30
Tue
Diffusion models for image generation II
  • INITNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR'24)
  • Multimodal Autoregressive Pre-training of Large Vision Encoders (CVPR'25)
  • Rethinking Cross-Modal Interaction in Multimodal Diffusion Transformers (ICCV'25)


  • TBD
  • TBD
  • TBD
10/2
Thr
Motion and video understanding I
  • Efficient Motion-Aware Video MLLM (CVPR'25)
  • Token Bottleneck: One Token to Remember Dynamics (arXiv'25)
  • Generative Omnimatte: Learning to Decompose Video into Layers (CVPR'25)


  • TBD
  • TBD
  • TBD
10/7
Tue
No Class (Korean Thanksgiving Day)
10/9
Thr
No Class (Korean Alphabet Day)
10/14
Tue
Diffusion for video generation and consistency models
  • 4Real-Video: Learning Generalizable Photo-Realistic 4D Video Diffusion (CVPR'25)
  • CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR'25)
  • Simplifying, Stabilizing and Scaling Continuous-time Consistency Models (ICLR'25)


  • TBD
  • TBD
  • TBD
10/16
Thr
Motion and video understanding II
  • Re-thinking Temporal Search for Long-Form Video Understanding (CVPR'25)
  • Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR'25)
  • V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (arXiv'25)


  • TBD
  • TBD
  • TBD
10/21
Tue
No Class (ICCV)
10/23
Thr
No Class (ICCV)
10/28
Tue
Text-to-Image (T2I) manipulation I
  • FreeControl: Training-Free Spatial Control of Any Text-to-Image Diffusion Model with Any Condition (CVPR'24)
  • Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML'24)
  • GenArtist: Multimodal LLM as an Agent for Unified Image Generation and Editing (NeurIPS'24)


  • TBD
  • TBD
  • TBD
10/30
Thr
Self-supervised multimodal modeling
  • Object-aware Sound Source Localization via Audio-Visual Scene Understanding (CVPR'25)
  • It's a (Blind) Match! Towards Vision-Language Correspondence without Parallel Data (CVPR'25)
  • Self-supervised spatial correspondence across modalities (CVPR'25)


  • TBD
  • TBD
  • TBD
11/4
Tue
Text-to-Image (T2I) manipulation II
  • UniReal: Universal Image Generation and Editing via Learning Real-world Dynamics (CVPR'25)
  • Unleashing In-context Learning of Autoregressive Models for Few-shot Image Manipulation (CVPR'25)
  • Concept Lancet: Image Editing with Compositional Representation Transplant (CVPR'25)


  • TBD
  • TBD
  • TBD
11/6
Thr
Multimodal representation learning
  • Continual Vision-Language Representation Learning with Off-Diagonal Information (ICML'23)
  • Emerging Properties in Unified Multimodal Pretraining (arXiv'25)
  • Seeing Speech and Sound: Distinguishing and Locating Audios in Visual Scenes (CVPR'25)


  • TBD
  • TBD
  • TBD
11/11
Tue
Flow models I
  • Stochastic Interpolants: A Unifying Framework for Flows and Diffusions (arXiv'23)
  • Inductive Moment Matching (ICML'25)
  • Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (ICML'24)


  • TBD
  • TBD
  • TBD
11/13
Thr
Vision-language foundation models
  • Understanding the Limits of Vision Language Models Through the Lens of the Binding Problem (NeurIPS'24)
  • Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR'25)
  • FLAIR: VLM with Fine-grained Language-informed Image Representations (CVPR'25)


  • TBD
  • TBD
  • TBD
11/18
Tue
Flow models II
  • Stochastic Optimal Control Matching (NeurIPS'24)
  • Align Your Flow: Scaling Continuous-Time Flow Map Distillation (arXiv'25)
  • Variational Rectified Flow Matching (ICML'25)


  • TBD
  • TBD
11/20
Thr
Analysis on multimodal LLMs
  • MLLMs Know Where to Look: Training-Free Perception of Small Visual Details with Multimodal LLMs (ICLR'25)
  • Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces (CVPR'25)
  • Do Vision and Language Encoders Represent the World Similarly? (CVPR'24)


  • TBD
  • TBD
  • TBD
11/25
Tue
Image generation with flow models
  • Improving Rectified Flow with Boundary Conditions (ICCV'25)
  • Normalizing Flows are Capable Generative Models (ICML'25)
  • STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (arXiv'25)


  • TBD
  • TBD
  • TBD
11/27
Thr
Object detection and segmentation
  • Efficient Test-time Adaptive Object Detection via Sensitivity-Guided Pruning (CVPR'25)
  • SAM 2: Segment Anything in Images and Videos (ICLR'25)
  • OMG-LLaVA: Bridging Image-level, Object-level, Pixel-level Reasoning and Understanding (NeurIPS'24)


  • TBD
  • TBD
  • TBD
12/2
Tue
Video generation with flow models and miscellaneous topics
  • Goku: Flow Based Video Generative Foundation Models (CVPR'25)
  • NFIG: Autoregressive Image Generation with Next-Frequency Prediction (arXiv'25)
  • Language-Guided Image Tokenization for Generation (CVPR'25)


  • TBD
  • TBD
  • TBD
12/4
Thr
Interpretability and miscellaneous topics
  • Cross-View Completion Models are Zero-shot Correspondence Estimators (CVPR'25)
  • Prompt-CAM: Making Vision Transformers Interpretable for Fine-Grained Analysis (CVPR'25)
  • CSA: Data-efficient Mapping of Unimodal Features to Multimodal Features (ICLR'25)


  • TBD
  • TBD
  • TBD
12/9
Tue
3D generation
  • Consistent Flow Distillation for Text-to-3D Generation (ICLR'25)
  • Neural Inverse Rendering from Propagating Light (CVPR'25)
  • DiffSplat: Repurposing Image Diffusion Models for Scalable 3D Gaussian Splat Generation (ICLR'25)


  • TBD
  • TBD
  • TBD
12/11
Thr
3D understanding
  • Point Transformer V3: Simpler, Faster, Stronger (CVPR'24)
  • Sapiens: Foundation for Human Vision Models (ECCV'24)
  • LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR'25)


  • TBD
  • TBD
  • TBD
Bldg 942-422, 1 Gwanak-ro, Gwanak-gu, Seoul 08826, Korea.
Copyright (c) 2021, Visual Information Processing Lab, Graduate School of Data Science, Seoul National University, All Rights Reserved.