SSD_Res50.md

PyTorch SSD_Res50

Environment Setup

Setup AI Model Efficiency Toolkit (AIMET)

Please install and setup AIMET before proceeding further. This model was tested with the torch_gpu variant of AIMET 1.24.

Install dependencies

   python -m pip install pycocotools gdown

We also need the original source as dependency for data processing purpose. You can clone the repo using the command:

   git clone https://github.com/uvipen/SSD-pytorch.git

Append the repo location to your PYTHONPATH with the following:

export PYTHONPATH=$PYTHONPATH:<path to SSD-pytorch>:<path to aimet-model-zoo>

Dataset

COCO 2017 val dataset can be downloaded from here:

• https://www.kaggle.com/datasets/awsaf49/coco-2017-dataset?select=coco2017

---

Usage

To run evaluation with QuantSim in AIMET, use the following

python3  aimet_zoo_torch/ssd_res50/evaluators/ssd_res50_quanteval.py \
                --model-config <configuration to be tested> \
                --dataset-path <path to the downloaded COCO dataset> \
                --use-cuda

Available model configurations are:

• ssd_res50_w8a8

---

Quantization Configuration

• Weight quantization: 8 bits, per tensor asymmetric quantization
• Bias parameters are not quantized
• Activation quantization: 8 bits, asymmetric quantization
• Model inputs are quantized
• TF was used as quantization scheme
• Cross-layer-Equalization have been applied on optimized checkpoint