ImgClsMob is an open-source sandbox and model zoo for training deep learning networks, best known for its large collection of image classification models implemented consistently across several frameworks. Rather than re-implementing the same architecture separately for every toolkit, ImgClsMob provides the same models โ ResNets, MobileNets, EfficientNets and many more โ in PyTorch, MXNet/Gluon, TensorFlow, Keras and Chainer, together with pretrained weights. The GitHub project has around 3,000 stars and was actively developed through 2024.
Comparing architectures or moving a model between frameworks usually means hunting down five different repositories with five different coding styles and training recipes. ImgClsMob removes that friction by keeping implementations in lockstep: one model definition style, shared training scripts, and checkpoints trained on standard datasets such as ImageNet and CIFAR. Beyond plain classification, the collection also covers related computer vision tasks, including semantic segmentation, human pose estimation and 3D face reconstruction, which makes it a handy reference implementation library for both research and teaching. Because the same model code targets several frameworks, it is also a practical Rosetta stone when migrating a vision project from one toolkit to another.
ImgClsMob is free and open source under the MIT license. There is no paid tier, cloud service or commercial edition โ you download the code and weights and run everything on your own hardware.
ImgClsMob is for computer vision researchers, students and engineers who want trustworthy reference implementations and pretrained checkpoints, especially when comparing architectures or porting models between deep learning frameworks.