Caffe Framework
AMD Vitis™ AI contains the following neural network libraries based on the Caffe framework:
TensorFlow Framework
Vitis AI contains the following neural network libraries based on the TensorFlow framework:
PyTorch Framework
Vitis AI supports the following type of neural network libraries based on the PyTorch framework.
- Classification
- ReID Detection
- Face Recognition
- Semantic Segmentation
- PointPillars
- Medical Segmentation
- 3D Segmentation
- PointPillars_nuscenes: Surround-view
- Centerpoint: 4D radar-based 3D detection
- PointPainting: Image-lidar sensor fusion
- Depth Estimation
- Bayesian Crowd Counting
- MultiTask V3
- Polyp Segmentation
- UltraFast Road Line Detection
- FairMot
- PSMNet
- SOLO
- CLOCs
- OCR
- Textmountain Detection
- Vehicle Classification
- OFA_YOLO Detection
- Monodepth2
- YOLOv5 Detection
- BEVDet Detection
- cFlownet
- YOLOv6 Detection
- YOLOv7 Detection
- YOLOv8 Detection
The related libraries are open-source and can be modified as needed. The source code is available on GitHub.
The Vitis AI Library provides test images and video test sequences for all the above networks. In addition, the Vitis AI Library package provides the corresponding performance test application. For video-based testing, use the raw video sequences for evaluation. The use of encoded video sequences for evaluation is not recommended as software decoders implemented on Arm processors may exhibit decode jitter which may affect the accuracy of evaluation.