imgclsmob

Why we included this project

If you build computer vision systems and often reuse pretrained backbones, this project saves you the usual boilerplate. It ships clean pip-installable packages for PyTorch, TensorFlow 1.x/2.x, Keras, MXNet/Gluon, and Chainer, each with a wide selection of classification, segmentation, detection, and pose estimation models. Pretrained weights for ImageNet, CIFAR, Cityscapes, COCO, and similar datasets load automatically when you instantiate a model, so you skip the download-and-map-to-weights step. Because the same architectures are mirrored across frameworks, it's also a convenient reference when you need to port a known model between ecosystems or verify that training behaves identically under different backends. The parent repository adds training and evaluation scripts if you want to go beyond loading a ready-made checkpoint.