-4

I’m working on a real-time system that identifies objects, but I’m facing a challenge: new objects can look extremely similar to known ones (sometimes differences are as small as 0.1 mm), and I need the system to detect when an object is truly new. Objects may appear in different positions and rotations, and I need to handle multiple objects in the same image.

Currently, I’m using YOLOv8, which works well for detecting and identifying pieces. However, since YOLO is primarily for detection and localization, I’m considering using a model like ResNet or VGG16 to extract visual features after YOLO detects each piece.

I’d love advice on whether this approach is considered good practice, or if there are better architectures or strategies for handling very similar objects while detecting unknown ones reliably.

Improve this question
4
  • 2
    Please provide enough code so others can better understand or reproduce the problem. Commented Nov 8 at 9:28
  • Hi Toni, I understand your point — unfortunately, I can’t share the full code since it’s quite complex and part of a larger industrial system. However, I can explain the core logic behind it. I’m working with very small aircraft parts that can look extremely similar (sometimes the differences are below 0.1 mm and barely visible). I use YOLOv8 for detection and localization, and each image comes from a 5400×3600 camera. Because of GPU constraints, I had to resize and crop the images carefully to keep as much detail as possible. Commented Nov 12 at 21:51
  • The dataset is also dynamically generated — images are captured and annotated automatically through code, so the full implementation is tightly coupled with the capture and annotation pipeline. That’s why I can’t easily isolate a “minimal reproducible example.” My main question is more architectural: whether using YOLO for detection and then a feature extractor (like ResNet or VGG16) for visual similarity and novelty detection is a good strategy for distinguishing between nearly identical parts and detecting unknown ones. Commented Nov 12 at 21:51
  • please take it to Cross Validated Commented Nov 12 at 22:55
Related questions
0
Yolo v8 slow fps in real time
0
Need advice for object detection and motion classification on real time video
1
Identifying moving objects using optical flow
Load 4 more related questions Show fewer related questions

0

Reset to default

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.