Image Communication and Understanding

Enhancing Video Object Detection Stability
Are you passionate about computer vision and eager to push the boundaries of video object detection? Join us in an exciting master thesis project focused on object detection in videos!

Challenge:
Traditional object detectors trained on images often struggle with instability when applied to videos. Motion blur, appearance changes, and frame-to-frame inconsistencies lead to unreliable detection, impacting the performance of many real-world applications.

Objective:
Develop a cutting-edge general adapter model that can be seamlessly integrated with existing image-trained object detectors. Your goal will be to enhance the stability and accuracy of object detection in videos, ensuring consistent results despite challenges such as motion blur and appearance variations.

Why This Project?

• Research Paper: Publish a paper at top-tier AI conferences such as CVPR/ICCV/ECCV.
• Real-World Impact: Improve the reliability of video-based applications such as augmented reality, and robotics.
• Technical Growth: Gain hands-on experience with state-of-the-art object detection models and advanced adaptation techniques.

Key Responsibilities:

• Review existing literature on image-trained object detectors and their performance on videos.
• Design and implement a general adapter model to stabilize detection results.
• Evaluate the model’s performance on benchmark video datasets.

Ideal Candidate:

• Strong background in computer vision, machine learning, and deep learning.
• Proficient in programming languages such as Python and frameworks like PyTorch.
• Analytical mindset with excellent problem-solving skills.
• Motivated, innovative, and eager to make a significant impact in the field.

Join us on this exciting journey to advance the capabilities of video object detection! Apply now and be a part of pioneering research that bridges the gap between image and video analysis.

Professor: Ender Konukoglu
Co-supervisor: Luc Van Gool

If you are interested in this project, please send your CV and transcripts to the co-supervisors:

Siyuan Li,
Dr. Guolei Sun,
Dr. Christos Sakaridis,

Feel free to contact us with any questions or to discuss the project further. We look forward to your application!
 

Project Description
Are you passionate about computer vision and 3D modeling? Do you want to be at the forefront of research in 3D point tracking? Join us in an exciting project where we aim to revolutionize 3D point tracking by creating a comprehensive synthetic dataset. The goal of this project is to develop a large-scale synthetic dataset tailored for 3D point tracking. This dataset will mimic real-world complexities, providing highly detailed and accurate 3D point annotations that are unattainable with current real-world datasets.

Project Scope

Data Generation

• Utilize advanced rendering techniques to generate highly realistic synthetic 3D environments.
• Incorporate diverse scene elements including various lighting conditions, dynamic objects, and natural movements of humans and animals.
• Leverage state-of-the-art tools like Blender for creating photorealistic renders..

Scene Diversity

• Generate scenes with a wide range of environments, from outdoor landscapes to intricate indoor settings.
• Include a variety of objects and textures to enhance the dataset’s richness and variability.

Learning Outcomes
Participating in this project offers several benefits:

• Gain hands-on experience with 3D modeling and rendering software.
• Develop a deep understanding of 3D point tracking and related computer vision techniques.
• Opportunity to publish in top AI conferences such as CVPR and ICCV.

Who Should Apply

• Students with a background in computer science or related fields.
• Individuals with experience in 3D modeling, rendering, or a strong interest in learning these skills.
• Enthusiastic learners who are eager to contribute to groundbreaking research and publish top-tier papers.
  

Professor: Ender Konukoglu
Co-supervisor: Luc Van Gool

If you are interested in this project, please contact the co-supervisors:

Siyuan Li,
Luigi Piccinelli,
Dr. Christos Sakaridis,

Are you passionate about computer vision and eager to push the boundaries of video segmentation in challenging scenes by exploiting vision foundation models? Join us in an exciting master thesis project that focuses on camouflaged object segmentation in videos!

Challenge
Camouflaged Video Object Segmentation (CVOD) aims to segment camouflaged objects that are seamlessly blended with their surroundings, as shown in the figure above. Existing methods often struggle with localizing camouflaged objects. This task has various real-world applications such as autonomous driving, medical video analysis, and wildlife protection.

Objective
Develop a cutting-edge CVOD model that can achieve state-of-the-art performance in existing benchmarks. Your approach will borrow knowledge from existing large vision foundation models such as GPT-4o or SAM to help with this task. Your goal is to enhance the stability and accuracy of camouflaged object segmentation in videos, despite the challenge that objects are blended into the background and thus difficult to observe.

Why this project?

• Research Paper: Publish a paper at top-tier AI conferences such as CVPR/ICCV/ECCV.
• Real-World Impact: Improve the reliability of video-based applications such as autonomous driving, and robotics.
• Technical Growth: Gain hands-on experience with state-of-the-art video segmentation models and large vision models.
  

Key responsibilities

• Review existing literature on CVOD, video segmentation, and vision foundation models (GPT-4o or SAM).
• Design and implement an advanced CVOD model.
• Evaluate the model’s performance on benchmark video datasets.
  
  

Ideal Candidate:

• Strong background in computer vision, machine learning, and deep learning.
• Proficient in programming languages such as Python and frameworks like PyTorch.
• Analytical mindset with excellent problem-solving skills.
• Motivated, innovative, and eager to make a significant impact in the field.

Join us on this exciting journey to advance the capabilities of video segmentation! Apply now and be a part of pioneering research that solves video segmentation under challenging scenes.
 

Professor: Ender Konukoglu

If you are interested in this project, please send your CV and transcripts to the co-supervisors:

• Dr. Guolei Sun,
• Siyuan Li,

Feel free to contact us with any questions or to discuss the project further. We look forward to your application!
 

Project Description
In the era of foundational models, machines are poised to replace or accelerate many tedious human tasks. However, training these foundational models requires massive amounts of high-quality data. Many computer vision tasks, such as detection, tracking, and video object segmentation, lack such data. In this project, we will explore and develop tools to utilize current foundational vision models, such as GPT-4o, SAM, etc., to perform automatic data annotation on videos.

The project’s goal is to develop the most efficient human and foundational model collaboration tools to minimize human efforts in obtaining high-quality annotated data at a low cost and publish a top-tier paper regarding your design.

Responsibilities
You will be provided with our existing codebase as a starting point. The basic functionalities and integration with GPT-4o and SAM are already in place. Your tasks will include:

• Enhancing the existing tool with additional foundational models.
• Improving the design for better human-model interaction.
• Deploying the tool into real production tests, conducting experiments, and refining features based on real human feedback.

Benefits
Participating in this project offers several benefits:

• Gain rich experience with various vision foundation models such as GPT-4o and assess their performance in real-world tasks.
• Acquire extensive software development and computer vision research experience.
• Opportunity to publish in top AI conferences such as CVPR and ICCV.

Candidate Profile
We are seeking a self-motivated student with a strong passion for learning and developing software with the latest AI models. The project requires full-stack software development skills. Familiarity with TypeScript and Go, the languages used in our codebase, is advantageous.

Professor: Ender Konukoglu
Co-supervisor: Luc Van Gool

If you are interested in this project, please contact the co-supervisors:

Siyuan Li,
Dr. Christos Sakaridis,