Vitis Video Analytics SDK Release V1.1¶

Summary¶

Based on Vivado 2021.2, Vitis 2021.2, Petalinux 2021.2
Based on Vitis AI 2.0 for Machine Learning
Supports:
- Zynq Ultrascale+ MPSoc based devices like ZCU104 Development Board and KV260 SOM Embedded Platforms
- Versal based platforms like vck190

Added support for Versal device based platforms
Support for cascaded Machine Learning
Introducing a new plug-in called vvas_xinfer, which supports:
- pre-processing (Resize, colorspace conversion, Normalization
- batch mode for increased DPU throughput
- Consolidating the inference information generated from previous levels into a single data structure.
Added support for two new models - Semantic Segmentation - Numberplate detection

Supports different input sources like Camera, RTP/RTSP streaming, file source etc.
H264/H265 Video Encoding/Decoding
Vitis AI based inferencing for detection and classification
Supported ML Models
- resnet50
- resnet18
- mobilenet_v2
- inception_v1
- ssd_adas_pruned_0_95
- ssd_traffic_pruned_0_9
- ssd_mobilenet_v2
- ssd_pedestrian_pruned_0_97
- plate_detect
- plat_num
- yolov3_voc_tf
- yolov3_adas_pruned_0_9
- refinedet_pruned_0_96
- yolov2_voc
- yolov2_voc_pruned_0_77
- densebox_320_320
- densebox_640_360
Hardware accelerated Resize and color space conversion
Region Of Interest based encoding
On-screen displaying bounding box around objects and text overly
HDMI Tx and Display Port interface for displaying the contents

For known limitations, refer to the example/application design for specific limitations, if any.

For known issues, refer to the example/application design for specific issues, if any.

We need one patch in XRT to fix an issue related to memory bank handling. This issue was seen in Versal based platform when plug-in tries to write into a buffer from memory Bank 3. Refer to XRT Patch for more details.