GenSynth Documentation

Model Checkpoint Packages

Model Checkpoint packages provide saved model checkpoints and associated Model Entities, along with full instructions on usage, for a variety of popular models for different tasks. The Model Checkpoint packages act as guides for how to prepare your own custom models in GenSynth for your development needs, and can even be used as the basis for your custom models with specific limitations on the format of the dataset. Available Model Checkpoint packages include:

Package Name

Description

Applies to:

avod_kitti_3d

Provides an untrained Aggregate View Object Detection (AVOD) checkpoint suitable to work with the KITTI Car 3D Object Detection (kitti_car_3d) dataset.

1.12+

centernet_resnet50_32x512x512

Provides an example checkpoint of the centernet-ResNet50 model.

1.13+

depthnet_kitti

Provides a DepthNet checkpoint with pre-trained weights. This model is compatible with KITTI Depth data.

1.12+

dncnn_128

A model checkpoint pacjage demonstrating the Denoising Convolutionsal Neural Network (DCNN).

1.14+

frcnn_resnet50_16x416x416

Provides a Faster-R-CNN (ResNet50) model, pre-trained on the VOC dataset. It requires input batch size 16 and input image 416x416x3.

1.12+

edge_speechnet_residual15

Provides an untrained checkpoint for Edge SpeechNet model.

1.11+

resnet50_224x224

Provides an untrained ResNet50 model that can identify 50 classes.

1.11+

simpnet_cifar10

Provides the SimpNet model checkpoint that has been pre-trained on the CIFAR10 dataset. This is part of the simpnet_tutorial.

1.11+

ssd_fpn_resnet50_16x512x512

Provides an untrained SSD-FPN (ResNet50) model, appropriate for use with the VOC 2007/2012 object detection dataset (voc0712_tfrecords).

1.11+

ssdlite_mobilenet_widerface

An example ssdlite_mobilenet model that is suitable for training to the widerface dataset.

1.13+

yolov3_16x416x416

Provides a pre-trained YOLOv3 model. It requires input batch size 16 and input image 416x416x3. It is also compatible with VOC0712 data.

1.12+