Capture-camera does not work

Hi @Sok_Phanit, the camera-capture tool saves your object detection dataset in Pascal VOC format, and not Open Images format. So when training model on your own custom data, you should use the --dataset-type=voc argument to train_ssd.py:

https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-collect-detection.md#training-your-model

Can you post the command line your are using to launch train_ssd.py?

Do you encounter an error when training on your own dataset, or does it not detect the objects? If the later, how many images are in your dataset?

When you run inferencing with detectnet program on your custom model, remember to use the labels.txt file that gets saved to your model directory (and not the labels.txt from your dataset directory). PyTorch adds a ‘background’ class to your labels during training and saves that to the labels.txt in your model’s directory.

Hello I try to train some images but I follow tutorials but it doesnt train. Please help

Hi @lkhagvasansar - which type of model are you trying to train? Classification or detection?

How many classes and images are in your dataset?

Does the training script produce an error or Python exception? If so, can you post console log and command line that you used to run the training script?

Thanks you @dusty_nv now, it’s working. I really want to training Segmentation with Hello AI on jetson nano , please make a guide for us.
My own training : My own training object detection base on (SSD-Mobilenet) Jetson nano - YouTube

And About Object detection API tensorflow 2.0 I try to training on jetson nano but I always get LOSS = nan.

First I convert XML file to csv and then to TFrecord for training. and I check path it’s didn’t wrong please give me some advice.
config file edit below :
Model : ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8
config file :
‘’‘num_classes: 4 ‘’’’

fine_tune_checkpoint: "/home/phanith/Desktop/object_detection/ssd_mobilenet_v2_fpnlite_320x320_c
fine_tune_checkpoint_type: “detection”

label_map_path: “/home/phanith/Desktop/object_detection/data/label_map1.pbtxt”

tf_record_input_reader
input_path: “/home/phanith/Desktop/object_detection/data/train.record”

label_map_path: “/home/phanith/Desktop/object_detection/data/label_map1.pbtxt”

tf_record_input_reader
input_path: "/home/phanith/Desktop/object_detection/data/test.record"

Hi @Sok_Phanit, Onixaz from the GitHub community has made a tutorial on training segmentation models, please see here -

I haven’t done TensorFlow training on Jetson myself, so you may want to create a new topic about it. Thanks.