AIHub Model Compilation, Inference, and Profiling Guide

This repository contains scripts to help you compile models, upload datasets, run inference, and profile models on AIHub. Follow the steps below to get started.

---

Prerequisites

Before running the scripts, ensure you have:

• A registered AIHub account
• Python 3.9+ installed
• The qai_hub Python SDK installed
• A trained model ready for compilation
• A dataset ready for upload

---

Steps to Use

Step 1: Registration

1. Go to AIHub's official website.
2. Create an account and set up your API credentials.

pip3 install qai-hub

# Navigate to Qualcomm® AI Hub website
# Account -> Settings -> API Token.
qai-hub configure --api_token INSERT_API_TOKEN

3. Go to lpcv.ai, locate Register button on homepage.
4. Register for participating in LPCVC 2025.

---

Step 2: Create Your Model

1. Train your own model

---

Step 3: Compile Your Model

1. Run the following script to submit your model for compilation:

python compile_model.py

Modify the compile_model.py script to:

• Replace model = "" with your actual model file.
• Set the target device based on your track.

The script will return a compiled job ID—save this for later steps.

For submission:

Go to lpcv.ai, locate Submit on homepage. Fill out the correct submission form for your track.

IMPORTANT: In order to submit the compiled model for evaluation, you must share it with the email address lowpowervision@gmail.com. To do this, modify the sharing settings in the script:

compile_job.modify_sharing(add_emails=['lowpowervision@gmail.com'])

This will grant the necessary permissions for the evaluation team to access the model. Sharing with this email is required for submitting your model for evaluation.

---

The following steps are for evaluting/testing your model on AI Hub:

Step 4: Upload Your Dataset

Run the following script to upload your dataset:

python upload_dataset.py

Modify the upload_dataset.py script to:

• Set image_folder = "Path/to/image/folder" to your dataset folder.
• Ensure the dataset consists of valid image files (.jpg, .png, .jpeg).

The script will return a dataset ID—save this for later steps.

---

Step 5: Run Inference

Once the model is compiled and dataset uploaded, run inference using:

python run_inference.py

Modify the run_inference.py script to:

• Set compiled_id with your compiled model ID.
• Set input_dataset with your dataset ID.

The script will submit an inference job.

---

Step 6: Profile Model Performance

To profile the compiled model, run:

python run_profile.py

Modify the run_profile.py script to:

• Set compiled_id with your compiled model ID.

This script will:

• Submit a profiling job.
• Print the estimated inference time upon completion.

---

Contact

For questions or issues, please use the lpcvc channel in the official Qualcomm® AI Hub slack channel.