Artificial Intelligence Models for Microplastic Detection

A repository presenting optimized neural networks for micro-plastic detection tasks.

How to Run the Experiments

Experiments Quantized MLP with generated C code

1 - Move to the repository directoty:

$ cd /micro-plastic/detection/repo/path

2 - Move to the directory of quantized MLP experiments:

$ cd experiments/C/qmlp/

3 - Make build.sh executable:

$ chmod +x build.sh

4 - Build the executable:

$ ./build.sh

5 - Run the executable to run the experiments:

$ ./run_exp_qmlp

This is an output running on architecture x86_64 Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz:

average: 0.00000043596000000023
max: 0.00007200000000000000

Experiments Quantized MLP with generated C code

1 - Move to the repository directoty:

$ cd /micro-plastic/detection/repo/path

2 - Move to the directory of quantized MLP experiments:

$ cd experiments/C/qgru/

3 - Make build.sh executable:

$ chmod +x build.sh

4 - Build the executable:

$ ./build.sh

5 - Run the executable to run the experiments:

$ ./qgru_benchmark

This is an output running on architecture x86_64 Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz:

average: 0.00000107914000000081
max: 0.00013600000000000000

Experiments Quantized MLP with ONNX Runtime

1 - Move to the repository directoty:

$ cd /micro-plastic/detection/repo/path

2 - Move to the directory of quantized MLP experiments:

$ cd experiments/python/qmlp/

3 - Move to the directory of quantized MLP experiments:

$ python framework_template.py

This is an output running on architecture x86_64 Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz:

average: 1.2655858993530274e-05
maximum: 0.0001380443572998047

Experiments Quantized GRU with ONNX Runtime

1 - Move to the repository directoty:

$ cd /micro-plastic/detection/repo/path

2 - Move to the directory of quantized GRU experiments:

$ cd experiments/python/qgru/

3 - Move to the directory of quantized MLP experiments:

$ python ort_gru_benchmark.py

This is an output running on architecture x86_64 Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz:

average: 3.22261381149292e-05
maximum: 0.00012826919555664062

Developers

Branch Name

Add your branch, with the following convention: [name]/[branch_name]

For example: alessandro/gemm_implementation

PR Labels

For PRs add one or more labels to classify the content:

• back-end: implementation of back-end in C.
• bug: fix a bug.
• documentation: adding documentation to the repository.
• experiments: code related to the experiments.
• front-end: implementation of models and training in Python.
• gru: code related to GRU model.
• mlp: code related to MLP model.

PR Assignee and Reviewers

Assign the PR to yuorself, if possible. It is possible to merge into the main branch without a review. Add someone as a review if you are unsure about the code.

Description

Add a short description of the PR content.