We are an automation company specializing in image processing solutions across various industries. Within our AI modules, we are leveraging Hailo AI technology to enhance our capabilities.
To optimize our projects, we require custom models that are compatible with Hailo modules. For instance, in our apple inspection project, we need an effective model that can accurately classify good and defective apples. Additionally, as we expand into different fruit inspection models, we recognize the necessity for specialized models tailored to specific applications.
We are writing to inquire if there are any outsourcing services available that specialize in the development of custom AI models compatible with Hailo modules
If you already have models but are not getting the desired accuracy, you could benefit from optimizing your models specifically for Hailo’s architecture. There are a few approaches you can take:
Model Optimization: You can use Hailo’s Model Zoo and Dataflow Compiler to optimize your models for Hailo’s hardware, which could help improve performance and accuracy.
Custom Model Development Services: While Hailo itself doesn’t offer direct outsourcing services for model development, there are third-party services and AI consulting firms that specialize in developing custom models optimized for specific hardware like Hailo. You can reach out to AI consulting firms with experience in edge AI deployments or companies specializing in Hailo’s technology.
If you are open to it, I would also suggest trying the Hailo Model Zoo, which includes pre-optimized models that may be adapted to your needs. Additionally, you can try tweaking your model’s training process or retraining the model with more domain-specific data.
Dear Omri,
I hope this message finds you well.
I am writing on behalf of Asaies Automation Pvt Ltd, an automation company based in India. We specialize in OEM electronics and control systems for machine manufacturing, focusing on developing various types of sorting machines for the past 25 years. Recently, we have embarked on developing an inspection module as an additional feature in our sorting machines.
To facilitate this, we have integrated your Hailo module with the Raspberry Pi 5 into our system. We successfully configured all settings and have been able to run your sample Python program, “hailo-rpi5-examples\basic_pipelines\detection.py,” with expected results.
However, we have encountered performance issues with the “instance_segmentation.py” example, where the segmentation processing runs significantly slower than anticipated. For our application, we require a performance of approximately 10-12 FPS at maximum resolution.
To address this issue, we plan to develop our own segmentation model tailored for fruit inspection to determine defect detection, size and color accurately. Given our experience with the Hailo module, we are reaching out for your support and guidance on the following:
Performance Optimization: What steps can we take to improve the FPS for instance segmentation while using the Hailo module with the Raspberry Pi 5?
Custom Model Development: Can you provide assistance or resources on how to create a custom segmentation model optimized for our specific use case?
Any insights or recommendations you could provide would be highly appreciated, as they will greatly assist us in moving forward with our project.
Thank you for your attention, and we look forward to your prompt response.
I’m glad to hear that you’ve successfully integrated the Hailo module with the Raspberry Pi 5.
Performance Optimization:
Achieving the performance you’re targeting, around 10-12 FPS, is definitely possible with some optimizations. Here are a few recommendations to help improve FPS:
Batch Size: Try running your model with a batch size of 8. This can improve throughput by processing multiple frames in one go.
Async Mode: Utilize asynchronous mode to run the model, which allows the system to handle multiple operations simultaneously, increasing the overall FPS.
Model Optimization: Optimizing the model for Hailo hardware will also help. You can do this by compiling the model with Hailo’s Dataflow Compiler (DFC) for better hardware utilization.
Custom Model Development:
If you’re looking to develop a custom model, I recommend visiting our Dataflow Compiler (DFC) guidelines at Hailo Developer Zone. This will guide you through optimizing your custom models specifically for Hailo hardware.
Additionally, please check out our ready-made instance segmentation models in the Hailo Model Zoo. These models are pre-optimized and may be adapted to your specific needs, potentially speeding up your development process.
Let me know if you need any further assistance or have more questions.