Hello everyone,
I’m working on a project in which I need to detect anomalies in a particular scene (two background scenes). The anomaly could be anything (bolts, pliers, glasses, etc.). However, I have generated synthetic data training with unity because I have very few realistic images and here comes the problem. I was looking throughout different techniques like Domain Adaptation since I need to find a solution that implies training on synthetic and doing inference on real ones, but each of them seems to be focused on the class of the image and thus could fail in finding particular anomaly. I’m not an expert in this field and I’d like to hear an expert’s opinion since I am a little bit stuck :(
thank you very much for you answers!
Refer this link
Article Using Imbalanced Triangle Synthetic Data for Machine Learnin...
I suggest you follow Chapter Privacy-Preserving Anomaly Detection Using Synthetic Data
https://cs.brown.edu/courses/cs227/archives/2017/papers/anomoly-survey.pdf
https://www.sciencedirect.com/topics/engineering/anomaly-detection
Giuseppe Giametta,
It all falls on how to get a representation space that resembles the underlying features of the real scene. I gather that you have knowledge of the realistic images on which you will test the data. Since, your objective with unity will be to generate a synthetic scene that resembles the one that you will be testing on. Doing this can be done if you are able to mimic the reals scene. This could be done via a complete synthetic reconstruction but the further away that you get the more difficult will be to achieve transferability. What you could do is a hybrid. You can insert scanned images of objects and textures that you would expect to identify in your real life scenario as a possible hybrid. The objective is to get as close as possible. Doing it completely synthetic will be a challenge to try to reproduce a scene unless you are a really good 3d artist.
The other alternative is to preprocess the data to make it look closer to a synthetic environment. This can be done by various filters that can be applied to the image.
Hope this helps
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training