Deep Learning Models and Tools
Deep Learning Models
- Darknet-19: Model and pre-trained weights
- Face detection using Tiny YoloV2: Face detection model and pre-trained weights
- Object detection using Tiny YoloV2: 313 class object detection model and pre-trained weights
- Object detection using YoloV2 Multisize: 313 class object detection model and pre-trained weights
- ResNet50: Network file, model parameter file, and layer mapping file for the convolutional neural network known in technical literature as ResNet-50.
- ResNet50_caffe: Model and pre-trained weights.
- ResNet101: Network file, model parameter file, and layer mapping file for the convolutional neural network known in technical literature as ResNet-101.
- ResNet152: Network file, model parameter file, and layer mapping file for the convolutional neural network known in technical literature as ResNet-152.
- ShuffleNetV2_sashdats: Model and pre-trained weights.
- U-Net: Network and model parameter files for the convolutional neural network known in technical literature as the U-Net.
- VGG-16: Network and model parameter files for the convolutional neural network known in technical literature as VGG-16.
- VGG-19: Network and model parameter files for the convolutional neural network known in technical literature as VGG-19.
- Xceptionweights: Model and pre-trained weights.
Deep Learning Example Data
- Create_Obj_Det_Train_Data: Source data set containing input image files and image annotation files for creating Object Detection model training data sets. This data is used with the SAS DLPy PrepareObjectDetectionTable example.
- convert_lenet_example: Support files for the example conversion of a Caffe LeNet model from BINARYPROTO to HDF5 format. Includes network file, model parameter file, and layer mapping file.
- fashion_mnist.zip: Training and test data for image classification using the Fashion MNIST data set.
- imagenet_example: Provides the files and instruction set to import a downloaded Caffe image classification model and evaluate its performance on a subset of the ImageNet data set. Includes a Jupyter notebook that contains a number of predefined CNNs, SAS code to evaluate a number of predefined CNNs, and a shell script to create a list of files and labels.
- image_segmentation_data_prep: Contains archived raw and mask image data sets for the SAS DLPy image segmentation data preparation example.
- image_semantic_segmentation_UNet_data: Contains the raw and annotated image data in .sashdat format used in the SAS DLPy semantic segmentation U-Net model example.
- Obj_Det_Soccer_Weights: Pretrained model weights used in the SAS DLPy FastRCNN object detection example.
- Obj_Det_Soccer_Images_416: Source data set for training and test data partitions used in the SAS DLPy FastRCNN object detection example.
- rnn_news_example: Support files for the example to build an RNN model using the news data set. Includes training data set, test data set, and word embeddings.
- speech_recognition_example: Support files for the DLPy audio training and speech recognition example notebook. Includes *.wav audio recordings, audio listing file, ground truth file, and input data cleansing file.
SAS Deep Learning Utilities
- caffe_model_conversion: Contains supplemental Caffe source code required to support model conversion from BINARYPROTO to HDF5 format.
- Model utilities: Provides Python code useful for importing all model definitions that SAS provides. Also provides Python function definitions and SAS Code definitions required for VGG-16, VGG-19, ResNet-50, ResNet-101, ResNet-152, and LeNet with Batch Normalization models.
Note:
These instructions and tools support importing BVLC Caffe models for use with SAS Deep Learning Actions.
SAS Computer Vision
- cv_example_data: Data set used in SAS Computer Vision examples that contains images of various size and type.