Jetson-inference detectnet

nicolasa.serrano · February 12, 2025, 3:06pm

I performed the detection of an object (Coin) with the detectnet example of jetson-inference, with 10 training images and 4 test images, but it always marks me twice in each image, does anyone know the reason? I used too few images? But it really is a very simple object.

nicolasa.serrano · February 12, 2025, 8:19pm

increases the sampling of 43 for training, 10 for testing and 11 for validation, it already identifies the correct amount of the object, but it identifies everything as currency, even wire, which I must modify in the training of this detection model.

AastaLLL · February 13, 2025, 4:36am

Hi,

Based on the documents below, we recommend at least 100 training images per class.

github.com/dusty-nv/jetson-inference

docs/pytorch-collect.md

master

<img src="https://github.com/dusty-nv/jetson-inference/raw/master/docs/images/deep-vision-header.jpg" width="100%">
<p align="right"><sup><a href="pytorch-plants.md">Back</a> | <a href="pytorch-ssd.md">Next</a> | </sup><a href="../README.md#hello-ai-world"><sup>Contents</sup></a>
<br/>
<sup>Transfer Learning - Classification</sup></s></p>

# Collecting your own Classification Datasets

In order to collect your own datasets for training customized models to classify objects or scenes of your choosing, we've created an easy-to-use tool called `camera-capture` for capturing and labeling images on your Jetson from live video:

<img src="https://github.com/dusty-nv/jetson-inference/raw/master/docs/images/pytorch-collection.jpg" >

The tool will create datasets with the following directory structure on disk:

```
‣ train/
	• class-A/
	• class-B/
	• ...
‣ val/
	• class-A/

This file has been truncated. show original

Could you increase the dataset and try it again?

Thanks.

nicolasa.serrano · February 20, 2025, 1:59pm

yes, I am using 380 images but now when I try to detect the same object at a distance greater than 30 centimeters it cannot locate any object.

any suggestions on this?

xplanescientist · February 21, 2025, 11:34am

Which Jetpack version are you running? 6.0 or earlier?

nicolasa.serrano · February 21, 2025, 1:06pm

Jetpack 6.0

nicolasa.serrano · February 21, 2025, 8:33pm

Now I have a problem with Docker, I don’t know what happened. It was working fine and I requested an update of the operating system and installed it, but since then I can’t run Docker.
I think it’s a problem with the network, something like a bridge.
I need your help to solve this.

AastaLLL · February 24, 2025, 8:27am

Hi,

The error is related to the recent docker 28.0.0 update.
You can find the WAR in the below comment:

Thanks.

nicolasa.serrano · February 24, 2025, 1:48pm

Thanks for the info, docker up date and run.

What is the reason that I do not detect the object at a distance greater than 30 centimeters, when I am using more than 300 images per class (Paste and NO_Paste).

This image is 20 cm from the objects:

and this is 35 centimeters away, from this distance it does not detect any object:

AastaLLL · February 25, 2025, 3:11am

Hi,

Are there some images with objects at a distance greater than 30 centimeters?

Please note that jetson-inference apply resizing before feeding the image to the DNN.
For example, SSD-Mobilenet input is 300x300 so it might not be good for the small objects.

There is discussion to increase the network input size for small objects:

Thanks.

nicolasa.serrano · February 28, 2025, 4:40pm

Yes, I have images larger than 30 centimeters that I used in training, I already changed the resolution to create the ssd-mobilente.onnx ($ python3 onnex_export.py --resolution=512 --model-dir=models/Coin/), and changed the threshold for the detectnet, these are my results, with the same image, which I actually used for training.
What modifications can I make to improve my model? I have already used more than 300 images for training, changed the resolution and performed tests with the threshold.

threshold=0.2

threshold= 0.21

threshold=0.22

threshold=0.23

threshold=0.24

threshold=0.25

threshold=0.26

threshold=0.27

threshold=0.28

threshold=0.29

AastaLLL · March 3, 2025, 6:52am

Hi,

Do you want to detect the circle or just the white square?
Could you also share some training images and the corresponding label with us?

Thanks.

nicolasa.serrano · March 5, 2025, 10:23pm

I want to detect the white square, I have two classes, when trying to do the training I have the following error.

AastaLLL · March 13, 2025, 3:26am

Hi,

The error comes from a custom script called Opencv_LabelSiemens.py.
It requires a library called tesseract.

Have you installed it in your environment?
If not, please try to install it and try it again.

Thanks.

Topic		Replies	Views
Problems with Jetson-inference Jetson Nano camera , jetson-inference , gstreamer	17	1252	November 30, 2022
Detect small Object Jetson Nano jetson-inference	4	129	September 2, 2024
Jetson nano start the Docker an error occurred while training your detection model ：Segmentation fault (core dumped) Jetson Nano jetson-inference	7	1232	April 21, 2022
Training model, object detetion Jetson Orin NX jetson-inference , ai-training , training , jetson	4	1050	September 26, 2023
Other Models for object detection Jetson Nano jetson-inference	18	1791	June 29, 2022
Train custom object detectio model Jetson Nano ai-training	12	3019	October 18, 2021
How train jetson-inference ssd512 model Jetson TX2 jetson-inference , ssd , pytorch	14	3017	October 18, 2021
Camera is showing part of my desktop instead of the image sometimes Jetson Nano camera	4	386	December 14, 2023
Custom Object Detection Model for Jetson nano using Deepstream DeepStream SDK cuda , tensorflow , ssd , yolo , nano	7	3521	November 23, 2021
Jetson nano detectnet Jetson Nano jetson-inference , python	12	870	June 2, 2023

Jetson-inference detectnet

Related topics