texasinstruments / edgeai-yolox

YOLOX-ti-lite models are optimized for deployment on TI edge processors

YOLOX based Models

This repository is a fork of YOLOX. This contains the enhancements of the YOLOX repository for supporting additional tasks and embedded friendly ti_lite models.

Installation

Step1. Install YOLOX.

./setup.sh

Step2. Install pycocotools.

pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

Tasks supported

• 2D Detection (with Ti-lite models)

  • These YOLOX based 2D detection models are optimized for TI processors.
  • Refer to this readme for further details.

• 6D Pose Estimation

  • 6D pose estimation or object pose estimation aims to estimate the 3D orientation and 3D translation of objects in a given environment. In this work, we propose a multi-object 6D pose estimation framework by enhancing the YOLOX object detector. The network is end-to-end trainable and detects each object along with its pose from a single RGB image without any additional post-processing.
  • Refer to this readme for further details.

• Keypoint Detection / Human Pose Estimation

  • Multi person 2D pose estimation is the task of understanding humans in an image. Given an input image, target is to detect each person and localize their body joints. In this work, we introduce a novel heatmap-free approach for joint detection, and 2D multi-person pose estimation in an image based on the popular YOLO object detection framework.
  • In general, this can be called Keypoint Detection or 2D Pose Estimation.
  • Refer to readme for further details.

Sample Inferences

• Given below are sample inferences for the tasks of human pose estimation and 6d pose estimation.

  Human Pose Estimation	6D Pose Estimation

Note:

See the original documentation

edgeai-yolox's Issues

Are you planning to add Yolov8 support?

6d pose onnx example models not working

Hi!

I am trying to run the demo code from https://github.com/TexasInstruments/edgeai-yolox/blob/main/README_6d_pose.md
using one of these models

and running

cd demo/ONNXRuntime
# Run inference on a set of sample images
python onnx_inference_object_pose.py --model "path_to_onnx_model"  --image-folder "path-to-input-images" --output-dir "ops_onnxrt" 

but the output folder with the bbox is always empty.

I am also trying to run this in onnx runtime directly but the result is always empty.

Would it be possible to say if the examples have to be setup in any way or the example code as extra requirements that are not explicit or what i could be doing wrong?

Thanks!

maxpool

Do you optimize the maxpool process as this optimizer figure.......

Dataset conversion from Yolo to COCO

Hello,

I am currently utilizing Yolov8-pose models to obtain predictions and corresponding labels in YOLO format for human pose estimation. However, I'm interested in finding out if there's a method to transform these labels from YOLO format into COCO format (.json file). This is because I aim to train Yolox-pose models using the annotations derived from the output of Yolov8-pose models.

where can i find yolox_s_human_pose_ti_lite.onnx moel for example test?

where can i find yolox_s_human_pose_ti_lite.onnx moel for example test? and pose test example code

About The Paper

Hello! I want to ask where is the link of the paper YOLO-6D-Pose: Enhancing YOLO for Multi Object 6D Pose Estimation which detailedly describes your algorithm?

Is possible to adapt the network for segmentation?

Hello, I would like to use YOLOX to train a segmentation model and deploy it on the TI board. Is it possible to modify the network to train a dedicated YOLOX model for the segmentation task?

TFLite export

Did anybody here try to export the onnx models to TFLite using onnx2tf?

I am running:

onnx2tf -i ~/edgeai-yolox/yolox_nano.onnx -o yolox_nano_saved_model

and get the following error

ERROR: input_onnx_file_path: /datadrive/mikel/edgeai-yolox/yolox_nano.onnx
ERROR: onnx_op_name: /head/ScatterND
ERROR: Read this and deal with it. https://github.com/PINTO0309/onnx2tf#parameter-replacement
ERROR: Alternatively, if the input OP has a dynamic dimension, use the -b or -ois option to rewrite it to a static shape and try again.
ERROR: If the input OP of ONNX before conversion is NHWC or an irregular channel arrangement other than NCHW, use the -kt or -kat option.
ERROR: Also, for models that include NonMaxSuppression in the post-processing, try the -onwdt option.

Any help is appreciated 😄