Hi everyone,
I’m currently working on my Final Year Project involving PCB defect detection using FOMO deployed on a Raspberry Pi 5.
I have successfully prepared my dataset and trained an initial model, but I am currently facing limitations with the Edge Impulse free plan that are affecting my progress.
Specifically:
- When increasing training cycles (e.g., 80–100), the job fails due to compute time limits
- Increasing image resolution also significantly increases training time beyond the allowed limit
- This makes it difficult to improve model performance for small PCB defects
I understand these limitations are part of the free tier, but for my project, I need to experiment with:
- Higher resolution inputs
- More training cycles
- Better optimization to improve detection accuracy
I have already submitted a request for the Enterprise Plan (for academic use), but I have not received any response so far.
I would really appreciate advice from the community:
- Is there any way to optimize training settings within the free plan to handle small object detection better?
- Are there recommended configurations to reduce compute time while maintaining accuracy?
- Has anyone successfully obtained Enterprise or academic access, and is there a recommended way to follow up or contact the Edge Impulse team?
This project is time-sensitive as part of my final year assessment, so any guidance would be extremely helpful.
Thank you in advance for your support!
Best regards,
Final Year Engineering Student