GluonCV

GluonCV is an open-source computer vision toolkit built on top of Apache MXNet’s Gluon API. It provides a curated collection of pre-trained models, model implementations, data loaders, and training utilities designed to accelerate research and production deployment in computer vision tasks.

The library features a comprehensive model zoo that covers multiple vision domains, including image classification, object detection, semantic and instance segmentation, pose estimation, and video action recognition. It includes well-known architectures such as ResNet, Inception, VGG, MobileNet, DenseNet, as well as detection and segmentation models like SSD, Faster R-CNN, RetinaNet, Mask R-CNN, DeepLab, and PSPNet. Pre-trained weights are available for common datasets, enabling quick fine-tuning and transfer learning.

GluonCV provides data handling and augmentation utilities, with dataset loaders for standard benchmarks such as ImageNet, COCO, PASCAL VOC, and Cityscapes. It also offers training and evaluation tools, including high-level training loops, distributed and multi-GPU support, mixed-precision training, learning rate schedules, and metrics for tasks across classification, detection, and segmentation.

For inference and deployment, GluonCV exposes simple APIs for running predictions, and supports exporting models to formats compatible with MXNet runtime and ONNX, facilitating deployment across CPUs, GPUs, and edge devices. The library emphasizes ease of use, clean API design, and extensibility, enabling researchers to experiment with architectures and pipelines while providing ready-to-use components for production-grade workflows.

GluonCV is maintained as an open-source project under the Apache 2.0 license, with contributions from individuals and organizations in the MXNet ecosystem. It serves as a community-driven resource for developing and evaluating computer vision models.