OpenVINO DL Workbench Problems

[Nowlan__Andrew1]

I have been having problems uploading data-sets to the workbench. The data-sets have all been in the image net format as described in the documentation. The work bench is recognizing the data-set to be of type 'ImageNet' but throws an error saying 'Cannot find annotation file. Any support with this issue would be greatly appreciated.

Thanks in advance. 

[JAIVIN_J_Intel]

Hi Andrew,

Thanks for reaching out.

Can you confirm that there is an annotation file in the dataset, which is organized as mentioned in the supported dataset formats ?

Regards,

Jaivin James

[Nowlan__Andrew1]

Hi Jaivin,

Thanks for the response. Yes, I have ensured the data-set includes an annotation file in the format described on your website. As a said in my previous post, the workbench recognises that the data-set is in the ImageNet format and throws an error ‘Annotation File Not found’. If I upload the same data-set without the annotation.txt file I get a different error: ‘Unknown dataset format’. So clearly, the file can be found. The workbench auto-generated data-sets load without issue but I am unable to locate these for a direct comparison. I have examined the DL Workbench Logs which provide a path to a directory where these auto-generated data-sets supposedly written to, however the specified directory does not appear to exist when i attempt to navigate to the specified location. 

Further assistance would be greatly appreciated.

Thanks in advance.

[JAIVIN_J_Intel]

Hi Andrew,

It seems like the issue is with the dataset structure. The auto-generated datasets can be found inside the docker container. You may navigate to the specified dataset directory after running the below command:

docker exec -it <CONTAINER ID> bash

To get the running container id:

docker ps

Feel free to ask any additional questions.

Regards,

Jaivin

[Sadmi]

Hi;

     If someone still facing this problem, in my case the content of the annotation file seems non valid as I used the classes names as labels IDs which is not accepted, so I just  changed the labels IDs with the numbers and it loaded as expected.

     I hope this helps.

Regards.