search

dusty-nv / jetson-inference

Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.
https://developer.nvidia.com/embedded/twodaystoademo
MIT License
7.86k stars 2.98k forks source link

How to make a Directory to train SSD-based Object Detection in PyTorch #776

Closed saifurrehman4114 closed 4 years ago

saifurrehman4114 commented 4 years ago

Hi, I have trained the model using the Camera Capture tool it was brilliant.

Now I am willing to train this model using my own data set which has about 100 Pictures of me not annotated yet,

  1. I am unable to figure first how to annotate them?
  2. How to make a directory structure as Camera Capture makes automatically to run this model?

If you could make a video it would be really helpful.

dusty-nv commented 4 years ago

Hi @saifurrehman4114 , I recommend to use the CVAT tool - https://github.com/openvinotoolkit/cvat

I tried it and it worked - I needed to make a labels.txt in my dataset, but otherwise it exported the data in Pascal-VOC structure.

I also needed to make a few minor tweaks today to the training script for it, so you should update or re-clone the repo. And if you are using the containers, run the docker/pull.sh script to get the latest container.

saifurrehman4114 commented 4 years ago

Thanks @dusty-nv. 1.Have you Tried LabelIMG tool to prepare data set?

  1. After using CVAT did you made JPEG folder to store images in them & train?
dusty-nv commented 4 years ago

Have you Tried LabelIMG tool to prepare data set?

Initially I tried LabelImg, but it did not create the proper Pascal-VOC directory structure - just the annotations.

  1. After using CVAT did you made JPEG folder to store images in them & train?

When you export your dataset from CVAT in Pascal VOC format, you will download a zip file that already has the JPEG folder, annotations folder, imagesets folder, ect. All you need to do is create a labels.txt for it, and put it inside your extracted dataset.

saifurrehman4114 commented 4 years ago

Thanks @dusty-nv. In CVAT when someone is the task Dashboard & someone tries to export Data Set in Pascal Format JPEG Folder doesn't come In my case working on their web application.

dusty-nv commented 4 years ago

Hmm, I notice that sometime, it can take a couple minutes for CVAT to prepare the zipped Pascal-VOC file for you to download (maybe longer if your dataset is long)

You can also try clicking on Tasks menu at top. Then Open your dataset, and try to export again from the Actions menu.

I believe you can also install CVAT tool locally on Linux/Windows/Mac.

saifurrehman4114 commented 4 years ago

Thanks @dusty-nv for the Guidance. Looking on creating something great.