Quantized and calibrated Resnet-50 v1.5 PyTorch model

Integer 8 bits precision weights for the Resnet-50 v1.5 PyTorch deep learning model. Created with the PyTorch post-training static quantization framework. PyTorch version 1.7.1 TorchVision version: 0.8.2 Quantization backend: FBGemm Activations: per channel asymmetric linear quantization (uint8_t datatype). Weights: symmetric linear quantization (int8_t datatype). The calibration data used during the quantization process consist on 500 images from the ImageNet validation dataset. The filenames list of these images can be found under: https://github.com/mlcommons/inference/blob/r1.0/calibration/ImageNet/cal_image_list_option_1.txt The base for this quantized model is the Resnet-50 32 bits precision PyTorch model provided by the TorchVision package. TorchVision package license: https://github.com/pytorch/vision/blob/master/LICENSE TorchVision package: https://github.com/pytorch/vision Original floating-point 32 bits precision PyTorch model hosted under the URL: https://download.pytorch.org/models/resnet50-19c8e357.pt