The Problem:
We have been diligently developing a model, utilizing both TensorFlow and Core ML for transfer learning. However, as we expand the number of classes, we've observed a decline in accuracy.

The Solution:
We are seeking an expert to address this challenge. The images, while visually distinct to a human observer, often have subtle differences that our model needs to detect accurately. Your expertise will be crucial in enhancing the model's capability to discern these nuances.

Once the model is working, we will need the ability to add classes as we scale up. Initially, the model should identify 100 classes and allow us to add up to 500 classes over time.

We can provide images for 50 classes for testing and optimization.

The Application:
The intended application of the model is on-device, necessitating a compact size suitable for seamless operation on a mobile phone. We are looking for a solution that not only improves accuracy with an increased class count but also ensures the model remains efficient for on-device use.

The Requirements:
1. Proven experience in optimizing models for on-device image classification/object detection.
2. Expertise in addressing accuracy decline issues when adding more classes to a model.
3. Familiarity with transfer learning using either TensorFlow or Core ML.
4. Ability to enhance model performance in detecting subtle differences in visually distinct images.
5. Experience in developing compact models suitable for mobile phone applications.

The Responsibilities:
1. Evaluate and troubleshoot the accuracy decline issue in our current model.
2. Optimize the model to accommodate an expanded class set while maintaining or improving accuracy.
3. Implement strategies to enhance the model's ability to detect subtle differences in visually distinct images.
4. Ensure the final model is compact enough for on-device deployment without sacrificing performance.
5. Develop a flexible architecture allowing the addition of classes, starting with 100 and scaling up to 500 over time.

The Deliverables:
1. Optimized image classification/object detection model with improved accuracy for an increased class count.
2. Documentation detailing the optimization process and parameters so we can continue our own ongoing model training.
3. Instructions for seamless integration of the optimized model into on-device applications.
4. Flexible model architecture enabling the addition of classes up to 500 as we scale up.

The Budget:
We welcome detailed proposals outlining the estimated budget for addressing the specified challenges and optimizing the on-device model. Include a breakdown of costs related to development, testing, and any additional considerations.

Let’s build something great:
Interested candidates are invited to submit proposals, highlighting relevant experience, strategies for addressing accuracy decline, and a proposed budget. Provide examples of previous work on optimizing on-device models if available.

Reference = Upwork , Link = https://www.upwork.com/freelance-jobs/apply/Image-Classification-Object-Detection-Expert-for-Device-Model-Optimization_~01a1abf8ceead1c0b6/