avatar

Chen-Yi Lu

Ph.D. Candidate
Purdue University
chenyilu777 (at) gmail.com

Download CV
On the job market Full-time, starting after May 2027

I am looking for industry roles as a:

Research Scientist Research Engineer Machine Learning Engineer
Email me View CV

About Me

I am a Ph.D. candidate at Purdue University in the ICAN Lab, advised by Prof. Somali Chaterji. I work on generative AI and multimodal models.

Diffusion models Video generation Vision-language models Multimodal LLMs Efficient on-device inference Adversarial robustness
Vision-language models
Fixing negation failures in CLIP with a small add-on that reads the frozen model's intermediate features (ECCV 2026 long oral).
Video generation
Decomposing finished videos into editable RGBA layers with a fine-tuned video diffusion transformer, and assembling coherent multi-shot videos from candidate clips (ICCV 2025).
Efficient inference
Training-free acceleration of text-to-image diffusion (ECCV 2024), single-step adversarial purification that runs in under 30 ms on NVIDIA Jetson, and benchmarking diffusion LLMs against autoregressive LLMs on edge hardware.

I have completed three research internships at Adobe Research (2023, 2024, 2026), working on efficient diffusion inference, multi-shot video creation and evaluation, and multimodal ad creative analysis with LLM agents. I am a co-inventor on one U.S. patent application.

Before Purdue, I was a Research Assistant at National Taiwan University (NTU) with Prof. Ta-Te Lin, where I built embedded vision systems and ML models for agricultural monitoring.

News

Jun 2026
Joining Adobe as a research intern for Summer 2026, working with Ishita Dasgupta and Stefano Petrangeli.
Jun 2026
Our paper What CLIP Knows but Cannot Say: Recovering Negation from Frozen Intermediate Features has been accepted to ECCV 2026 as a long oral (28 of 2,864 accepted papers)! See the project page.
Oct 2025
Co-organized the 1st Workshop on Short-Form Video Understanding (SVU) at ICCV 2025.
Mar 2025
Our paper SKALD has been accepted to ICCV 2025!
Feb 2025
Our paper on semi-supervised semantic segmentation accepted to CVPR 2025!
Older news
Jul 2024
Our paper ReCon has been accepted to ECCV 2024!
May 2024
Returned to Adobe Research for a second summer internship.
Aug 2023
Started Ph.D. at Purdue University.
May 2023
Started summer internship at Adobe Research.

Selected Publications All publications ยท Google Scholar

  1. ECCV Oral
    What CLIP Knows but Cannot Say: Recovering Negation from Frozen Intermediate Features
    Chen-Yi Lu, Yueh-Shao Chen, Somali Chaterji
    European Conference on Computer Vision (ECCV), 2026. Long oral (top 1% of accepted papers)
    Project Page arXiv PDF
  2. ICCV
    SKALD: Learning-Based Shot Assembly for Coherent Multi-Shot Video Creation
    Chen-Yi Lu, Md Mehrab Tanjim, Ishita Dasgupta, Somdeb Sarkhel, Gang Wu, Saayan Mitra, Somali Chaterji
    International Conference on Computer Vision, 2025.
  3. CVPR
    Improving Semi-supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity
    Chen-Yi Lu, Kasra Derakhshandeh, Somali Chaterji
    The IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025.
  4. ECCV
    ReCon: Training-Free Acceleration for Text-to-Image Synthesis with Retrieval of Concept Prompt Trajectories
    Chen-Yi Lu*, Shubham Agarwal*, Md Mehrab Tanjim, Kanak Mahadik, Anup Rao, Subrata Mitra, Shiv Kumar Saini, Saurabh Bagchi, Somali Chaterji
    European Conference on Computer Vision (ECCV), 2024. *Equal contribution.

Service


Powered by Jekyll and Minimal Light theme.