egonet

EgoNet

Input

(Image from http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d)

Output

• red: use detected bbox
• yellow: use ground truth bbox

Data Preparation

You need to download KITTI dataset here. Download left images, calibration files and labels. Also download the D4LCN.zip (predicted).
Data folder should look like this:

egonet
├── calib
  ├── xxx.txt (Camera parameters for image xxx: provided from data_object_calib.zip)
├── label
  ├── xxx.txt (predicted object labels for image xxx: provided from D4LCN.zip)
├── gt_label
  ├── xxx.txt (ground-truth object labels for image xxx: provided from data_object_calib.zip)

These files are written in kitti format.

Usage

Automatically downloads the onnx and prototxt files on the first run. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 egonet.py

If you want to specify the input image, put the image path after the --input option.
You can use --savepath option to change the name of the output file to save.

$ python3 egonet.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

You can specify directory or directly file path with the --label_path, --gt_label_path and --calib_path options for find label, gt_label, calib file.

$ python3 egonet.py --label_path LABEL_PATH --gt_label_path GT_LABEL_PATH --calib_path CALIB_PATH

You can get the 3D plot by specifying the --plot_3d option.

$ python3 egonet.py --plot_3d

For images without a label file, you can use the --detector option to get the BBOX.

$ python3 egonet.py --detector

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 egonet.py --video VIDEO_PATH

Reference

• EgoNet

Framework

Pytorch

Model Format

ONNX opset=11

Netron

HC.onnx.prototxt
L.onnx.prototxt