Map near to 0

Accelerated Computing Intelligent Video Analytics TAO Toolkit
,
1

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) t4
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) Detectnet_v2
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) don’t have it
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)
detectnet_v2_train_custom.txt (3.3 KB)
status.txt (52.6 KB)

hi can someone help me in this?
i’m trying to use detectnet_v2 as my model in tao but i got mapp near to 0 without changing in my loss
can you help me?
my data has different size but i use auto_size so it’s not the problem
my class terget is correct ! the same as my label files in kitti format

2

Could you please try with 4.0.1 docker? Thanks.
$ docker pull nvcr.io/nvidia/tao/tao-toolkit:4.0.1-tf1.15.5

3

which version of notebook i have to use for this ? does it work with 5.3 in notebook too?

4

Yes, you can use 5.3 notebook.

1 Like
5

i pull it and remove my last image but it automaticly try to isntall tao-toolkit 5.0.0-tf1.15.5

6

Please use docker run way instead of tao-launcher.
Refer to Getting 0 mAP for detectnet_v2 model over 150 epochs - #6 by Morganh.

7

Do you mean you set enable_auto_resize: true?

May I know that if you can run default notebook and default dataset successfully?

8

Is there any guide for it?

9

You can see my config file so it must work( i think )
The dataset is to large so i try to don’t use it

10

I don’t understand correctly what is this?
In the second command what is /localhost/marganh ? And Why it has /bin/bash at the end
In the third command : notebook have some envierment to set before running tao model train … how can i handle that here?

I mean how to say it where is my workspace ?(in python notebook there is some variables that must be set in notebook forexample where is my workspace that specs and dataset exist there)

11

The notebook tells user to install tao-launcher. Instead, I am suggesting you to run with docker run instead of tao-launcher. The -v is in order to map your local files to docker. This helps you to mount your training dataset/spec file into docker. It is a common usage of docker. You can search the docker doc to take look.
For the environment you set in notebook, you can ignore now because you will run with docker run in order to run inside the docker.
For the enable_auto_resize: true, please see in the doc https://docs.nvidia.com/tao/tao-toolkit/text/cv_finetuning/tensorflow_1/object_detection/detectnet_v2.html.