Vision AI weekly: Issue 01

Another exciting week in the Vision AI ecosystem!

Author

sriram Venkatesan
July 22, 2025 • Estimated Reading Time: 4 minutes

🌟 Editor's Note
Welcome to another exciting week in the Vision AI ecosystem! We've got a packed newsletter full of insights, events, and inspiring stories from the heart of innovation.

🗓️ Event Highlights

Embodied Computer Vision at CVPR 2025:

The CVPR 2025 conference highlighted Embodied Computer Vision as the next AI frontier, transitioning from passive perception to intelligent, context-aware action. Talks showcased advancements like RoBoSpatial for spatial reasoning, GROVE for generalized reward learning, and Navigation World Models for planning in simulation. Google DeepMind's Gemini Robotics demonstrated bridging foundation models with physical actions, underscoring that embodied AI is here, capable of interacting, learning, and evolving within the physical world. [link]

Deadlines:

  • AAAI’26 (paper) : 12 days

  • 3DV’26 : 28 days

  • WACV’26 (R2, reg) : 54 days

  • WACV’26 (R2, paper) : 61 days

🚀 Case study

The Rise of Vision AI in aerospace

Visual automation, utilizing AI and computer vision, is transforming aerospace quality control. It offers faster, more accurate inspection than manual methods, crucial for complex aircraft manufacturing. Tools like no-code platforms make AI accessible, enabling real-time defect detection, auditable records, and scalability, ultimately enhancing safety and compliance in the aerospace industry. [link]

🦄 Startup Spotlight

36ZERO Vision: AI- powered visual inspection for manufacturing

36ZERO Vision, a Munich-headquartered startup specializing in AI-powered visual inspection for manufacturing, has secured €3.6 million in pre-Series A funding. This investment, led by JOIN Capital, Bayern Kapital, and Vanagon Ventures, will facilitate the expansion of its AI visual inspection platform.

The company's proprietary multi-stage deep learning model requires as few as five annotated images, significantly outperforming competitors by reducing pseudo-defects and accurately simulating thousands of variations. 36ZERO Vision's software-defined, hardware-agnostic technology boasts superior accuracy and ease of use, leading to a remarkable reduction in false positives.

The new capital will be used to scale operations across Europe, expand with existing customers like Siemens and Bosch Rexroth, and enhance its team and product development. The company aims to further strengthen its leadership in the global AI-driven visual inspection market.

🔥 Paper to Factory

This paper introduces SWIM, a new dataset of nearly 64k annotated spacecraft images, designed for real-time segmentation on onboard flight computers. It addresses the scarcity of public data by combining real spacecraft models with real and synthetic backgrounds, including noise and distortions. The authors finetuned YOLOv8 and YOLOv11 models, achieving a Dice score of 0.92 and Hausdorff distance of 0.69, with an inference time of approximately 0.5 seconds, meeting NASA's hardware constraints.

🏆 Community Spotlight:

  • In the recent Voxel 51 podcast, Jason Corso highlights the rise of Visual AI applications in manufacturing

  • In the latest Hansel minutes podcast, Sky Engine AI’s Dr. Marc Scouter dives deep into the surprising power of synthetic data, exploring when fake can outperform real in areas like medical imaging, defense and self driving cars

  • In their latest Visual AI in manufacturing report, Voxel 51 discusses the future of AI in manufacturing, the leading companies and voices driving innovation in the space

  • Roboflow’s latest tutorial highlights on how to annotate a dataset and train a vision AI model

Reddit / X corner:

  • Ultralytics discuss on how to detect rotated objects with Ultralytics YOLO OBB task and how to train YOLO11 on the LVIS dataset for long-tail object detection

  • A latest Reddit post discusses on problems in training EfficientDet model for Edge TPU

  • Another reddit post discusses on training a segmentation model when an object has optional parts and annotations are inconsistent

Till next time,

Vision AI weekly