DSP block error, wont process img data

Question/Issue:
Hello
I use a small jupyter notebook to upload a MNIST dataset of handsigns.
The dataset is a csv file containing around 34.000 datapoints which each contain 784 samples, 28x28 pixels.
I can upload them to Edge Impulse Studio, however when i go into the image preprocessing block the DSP result outputs the following:
Error: The specified key does not exist.

This is how my json looks like:
{‘protected’: {‘ver’: ‘v1’, ‘alg’: ‘HS256’, ‘iat’: 1663877227.903796}, ‘signature’: ‘0000000000000000000000000000000000000000000000000000000000000000’, ‘payload’: {‘device_type’: ‘pre-made’, ‘interval_ms’: 100, ‘sensors’: [{‘name’: ‘img’, ‘units’: ‘B’}], ‘values’: [107.0, 118.0, 127.0, 134.0, 139.0, 143.0, 146.0, 150.0, 153.0, 156.0, 158.0, 160.0, 163.0, 165.0, 159.0, 166.0, 168.0, 170.0]}}

What am i missing ?

Project ID:
139321
Context/Use case:
Training an image classifier, following the Edge Impulse tutorial.

Hello @dkhusted,

I think your issue come from the image format which is not accepted by Edge Impulse. For images, we only work with .jpg or .png.

In that case, I would create a simple Python script that convert the csv rows to png/jpg image and use our API/CLI to upload the images to your project.

Best,

Louis

Hey @louis
Thanks for the reply!
Yea i figured that out as well after some time, and did end up making a Python script which convert the whole CSV into to png, which i uploaded through the API with great succes!
Are there any plans to expand the ingestion service to accept imgs in csv format in the long run? As in my case 34.000 png’s did take abit of time(around 3 hours), as the uploader started to slow down around 4.000 png files in. Could be an restriction from my side of course.

Hello @dkhusted,

No, not in the sort term at least…

Our enterprise customers can use Transformation blocks that can automatically grab a csv file in a cloud bucket, parse it and import it in a project.

Best,

Louis