Please provide the following information when requesting support.
• Hardware 1070ti
• Network Type Detectnet_v2/
• TLT Version 3.x
Error when using tao tool to train detectnet_v2 detection model
Not found: /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images/
/home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images/smoke_17982.png; No such file or directory
There is no png file in my dataset, I don’t know what is wrong, I have completed the tf format conversion
tensorflow.python.framework.errors_impl.NotFoundError: 2 root error(s) found.
(0) Not found: /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images//home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images/smoke_17982.jpg; No such file or directory
Many topics are asking this.
Please check the your ~/.tao_mounts.json.
The path xxx should be a path inside the docker when your run command "tao detectnet_v2 xxx "
Can you run below and share the log?
$ tao detectnet_v2 run ls /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/specs/detectnet_v2_train_resnet18_kitti.txt
The problem I found is that he automatically added an images directory after the directory, but I didn’t add it, the system automatically added an extra images directory
Not found: /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images/images/smoke_25325.jpg
According to above error log, the image path is not available.
Please run below command to login the docker and then debug by yourself.
$ tao detectnet_v2 run /bin/bash # detectnet_v2 train -k nvidia -r /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/result/ -e /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/specs/detectnet_v2_train_resnet18_kitti.txt
still so
Not found: /home/ncp/tao/cv_samples_v1.2.0/detectnet_v2/dataset/images/images/smoke_25325.jpg; No such file or directory
You look at my config file above, isn’t it there?
Hey,
I suggest you to run all steps inside the docker for generating tfrecords and training.
$ tao detectnet_v2 run /bin/bash (login the docker)
Then inside the docker, firstly, please make sure all your dataset are available. @6a4b7055d229:/workspace$ ls /home/ncp/tao/cv_samples_v1.2.0/dataset/images/ (make sure you find your dataset inside the docker)
yes am sure，you can take a look at this screenshot of mine, the last time I ran the training, it was possible to start training, but as soon as it was started, it was killed directly
also, I can upload a few training images, label files and configuration files for you, can you help me?
If you already find your training images, please try to generate the tfrecord files again inside the docker.
I am afraid there is something wrong in this step.
