All Projects → TexasInstruments → Jacinto Ai Devkit

TexasInstruments / Jacinto Ai Devkit

Licence: other
Training & Quantization of embedded friendly Deep Learning / Machine Learning / Computer Vision models

Projects that are alternatives of or similar to Jacinto Ai Devkit

Cnn Paper2
🎨 🎨 深度学习 卷积神经网络教程 :图像识别,目标检测,语义分割,实例分割,人脸识别,神经风格转换,GAN等🎨🎨
Stars: ✭ 77 (+57.14%)
Mutual labels:  object-detection, cnn, deeplearning, segmentation, detection
PaddlePaddle End-to-End Development Toolkit(『飞桨』深度学习全流程开发工具)
Stars: ✭ 3,399 (+6836.73%)
Mutual labels:  deeplearning, segmentation, resnet, mobilenet, detection
Caffe Model
Caffe models (including classification, detection and segmentation) and deploy files for famouse networks
Stars: ✭ 1,258 (+2467.35%)
Mutual labels:  segmentation, caffe, resnet, detection
Model Quantization
Collections of model quantization algorithms
Stars: ✭ 118 (+140.82%)
Mutual labels:  cnn, segmentation, quantization, detection
Tools To Design Or Visualize Architecture Of Neural Network
Tools to Design or Visualize Architecture of Neural Network
Stars: ✭ 1,143 (+2232.65%)
Mutual labels:  cnn, deeplearning, semantic-segmentation, resnet
The Medical Detection Toolkit contains 2D + 3D implementations of prevalent object detectors such as Mask R-CNN, Retina Net, Retina U-Net, as well as a training and inference framework focused on dealing with medical images.
Stars: ✭ 917 (+1771.43%)
Mutual labels:  object-detection, segmentation, semantic-segmentation, detection
This repository contains the source code of our work on designing efficient CNNs for computer vision
Stars: ✭ 331 (+575.51%)
Mutual labels:  object-detection, cnn, semantic-segmentation
Php Opencv Examples
Tutorial for computer vision and machine learning in PHP 7/8 by opencv (installation + examples + documentation)
Stars: ✭ 333 (+579.59%)
Mutual labels:  caffe, mobilenet, detection
Rectlabel Support
RectLabel - An image annotation tool to label images for bounding box object detection and segmentation.
Stars: ✭ 338 (+589.8%)
Mutual labels:  object-detection, segmentation, detection
High level network definitions with pre-trained weights in TensorFlow
Stars: ✭ 982 (+1904.08%)
Mutual labels:  object-detection, resnet, mobilenet
Segmentation models
Segmentation models with pretrained backbones. Keras and TensorFlow Keras.
Stars: ✭ 3,575 (+7195.92%)
Mutual labels:  segmentation, resnet, mobilenet
Tf Faster Rcnn
Tensorflow Faster RCNN for Object Detection
Stars: ✭ 3,604 (+7255.1%)
Mutual labels:  object-detection, resnet, mobilenet
LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation, CVPR 2018 (Spotlight paper, 6.6%)
Stars: ✭ 474 (+867.35%)
Mutual labels:  cnn, deeplearning, caffe
Yolo V2 Pytorch
YOLO for object detection tasks
Stars: ✭ 302 (+516.33%)
Mutual labels:  object-detection, cnn, deeplearning
Segmentation keras
DilatedNet in Keras for image segmentation
Stars: ✭ 300 (+512.24%)
Mutual labels:  deeplearning, segmentation, caffe
Fast image augmentation library and an easy-to-use wrapper around other libraries. Documentation: Paper about the library:
Stars: ✭ 9,353 (+18987.76%)
Mutual labels:  object-detection, segmentation, detection
Cascaded Fcn
Source code for the MICCAI 2016 Paper "Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional NeuralNetworks and 3D Conditional Random Fields"
Stars: ✭ 296 (+504.08%)
Mutual labels:  segmentation, semantic-segmentation, caffe
Jetson Inference
Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.
Stars: ✭ 5,191 (+10493.88%)
Mutual labels:  object-detection, segmentation, caffe
Tensorflow Face Detection
A mobilenet SSD based face detector, powered by tensorflow object detection api, trained by WIDERFACE dataset.
Stars: ✭ 711 (+1351.02%)
Mutual labels:  object-detection, mobilenet, detection
LightNet: Light-weight Networks for Semantic Image Segmentation (Cityscapes and Mapillary Vistas Dataset)
Stars: ✭ 698 (+1324.49%)
Mutual labels:  segmentation, semantic-segmentation, mobilenet

Jacinto AI DevKit

Training & Quantization Tools, Model Zoo & Accuracy Benchmarks For Embedded AI Development.



Embedded Inference of Deep Learning models is quite challenging - due to high compute requirements. This repository provides various tools for AI, including Deep Learning, traditional Machine Learning and Computer Vision. We show several low complexity Deep Learning models that make Inference on Low Power Embedded Platforms practical. We also have examples for Quantization Aware Training for 8-bit fixed point inference.

Several of the Deep Learning models described in these repositories have been verified to work on TI's Jacinto Automotive Processors using TI Deep Learning Library (TIDL) that is part of Processor SDK RTOS for Jacinto7.

These tools and software are primarily intended as examples for learning and research.

Each repository listed below has a documentation page and a separate summary page containing git clone/pull URLs. Please browse those links to read the documentation as well as to get the git URLs for cloning the repositories. This github repository does not contain any code, but has links that point to the actual repositories containing code.

Model Zoo & Accuracy Benchmark

  • Jacinto-AI-ModelZoo: Large collection of pre-trained models that are verified to work on our platform. We also information on how to train some of the important models, and also the complexity and what accuracy to expect. (coming soon...).
  • Jacinto-AI-Benchmark: Accuracy benchmarking of Deep Learning models is a difficult task. We make it easy for our platform with a Python package that runs on PC as well as on our EVM. The pre-trained models in our Model Zoo are supported off-the-shelf in this benchmark code. Accuracy benchmark of custom models can also be done easily with just a few lines of code. (coming soon...).

Training Repositories

Deep Learning Examples for Jacinto7 family of devices - e.g. TDA4VM

  • Pytorch-Jacinto-AI-DevKit: PyTorch based training of Image Classification, Semantic Segmentation, Depth Estimation, Motion Segmentation and various other Pixel2Pixel tasks, Multi-Task Learning and Quantization Aware Training.
    -- Documentation: link
    -- git clone/pull URLs: link

  • PyTorch-MMDetection: PyTorch based Object Detection training/quantization
    -- Documentation: link
    -- git clone/pull URLs: link

  • TensorFlow2.0 based quantization examples - (coming later...).

Deep Learning and Traditional ML examples for Jacinto 6 family of devices - e.g. (TDA2x, TDA3x). These older modules are not included as submodules in this repo, but can be obtained using the links below.

  • Caffe-Jacinto: Our Caffe fork for training sparse CNN models including Object Detection and Semantic Segmentation models.
    -- Documentation: link
    -- git clone/pull URLs: link

  • Caffe-Jacinto-Models: Scripts and examples for training sparse CNN models for Image Classification, Object Detection and Semantic Segmentation.
    -- Documentation: link
    -- git clone/pull URLs: link

  • Acf-Jacinto - training tool for HOG/ACF/AdaBoost Object Detector (traditional machine learning based)
    -- Documentation: link
    -- git clone/pull URLs: link

Model Quantization

Quantization (especially 8-bit Quantization) is important to get best throughput for inference. Quantization can be done using either Post Training Quantization (PTQ) or Quantization Aware Training (QAT).

TI Deep Learning Library (TIDL) that is part of the Processor SDK RTOS for Jacinto7 natively supports PTQ - TIDL can take floating point models and can quantize them using advanced calibration methods.

We have guidelines on how to choose models and how train them to get best accuracy with Quantization. It is unlikely that there will be significant accuracy drop with PTQ if these guidelines are followed. In spite of this, if there are models that have significant accuracy drop with quantization, it is possible to improve the accuracy using QAT. Please read more details in the pytorch-jacinto-ai-devkit and its documentation on Quantization.


  • If you click on one of the links above - it will navigate to a repository hosted in From there, you can click on one of the tabs to get more information on that repository.
  • For example, the about tab shows documentation. The summary tab lists all the branches, commit information and links for git clone/pull that repository.
  • Each of those repositories also have separate LICENSE files.

What is New:

  • [2020-August-05] pytorch-mmdetection is made available with several training/quantization/evaluation examples for object detection. See the links for more information.
  • [2020-August-05]: We do not use git submodule anymore. You can clone or update the repositories mentioned in the Repositories section below by git clone or git pull.
  • [2020-June-15]: Our internal URLs changed. If you cloned the repository before this date please do a fresh clone to avoid confusion. Without this, doing a ./ may appear to succeed, but may not be actually updating the submodules.
  • [2020-January-28] pytorch-jacinto-ai-devkit: Documentation and results for Depth Estimation has been added.
  • [2020-January-06] pytorch-jacinto-ai-devkit: Quantization documentation is updated. Trained Quantization using QuantTrainModule is now the recommended method of doing Quantization Aware Training and the use of QuantCalibrateModule is discouraged.


Please see the LICENSE file for more information about the license under which this landing repository is made available. The LICENSE file of each repository mentioned here is inside that repository.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].