Files
SDK_SG200x_V2/cviruntime/samples/samples_extra
carbon e25f20f7a3 add cviruntime
commit 3f4938648950a7f3bf9a19c320ca9fae7c52de20
Author: sophgo-forum-service <forum_service@sophgo.com>
Date:   Mon May 13 13:44:23 2024 +0800

    [feat] cviruntime opensource for cv18xx soc.

    - a4b6a3, add cumsum and gatherelements_pt.
2024-05-31 11:51:34 +08:00
..
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00
2024-05-31 11:51:34 +08:00

Samples extra for CVI TPU SDK

Yolo detection series

The sample implementation of yolov3 without post-processing, yolov5 with post-processing and yolox is provided. Please refer to the detector_yolov3_fused_ preprocess, detector_yolov5_fused_Preprocess and detector_yolox_s Implementation under directory

Preprocess classification

Support the use of TPU or VPSS for pre-processing. Take the classification model as an example:

  1. Refer to classifier for classifier_tpu_preprocess under preprocess directory
    The advantage is to reduce the memory copy of pre-processing, but it will increase the use of ion

  2. Refer to classifier for classifier_vpss_preprocess under preprocess directory
    The advantage is that VPSS supports more types of preprocessing. It does not need to use TPU for additional preprocessing, but memory copying is required

Complex deployment scenario sample

  1. for attitude evaluation, refer to alphapose_fused_preprocess directory
  2. for face recognition detection, please refer to insightface_fused_preprocess directory