Issue in image segregate in "Image" section of edge impulse

Report bugs
#1

Question/Issue:
Facing problem with labeled images

Project ID:
684050

Context/Use case:
Thi issue that I am encountering is I am trying to train an object detection model with animals. While doing the labeling and tuining, I found out that the platform is unable to segregate between 2 animals and in “Image” section, ‘feature explorer’ I see a compeletly overlaping issue. So I added a third animal to see if anything changes, but no, still the same. My feature explorer looks kind of all messed up, have a look below. I am not understanding why this is happening.

Capture
Capture464×522 75.3 KB

For one instance I can understand zebra and tiger can be a misleading animals due to the black straps they have, but elephant no chance, it doesn’t have a single common characteristic as the remaining 2 animals.

Please someone guide me on this.

Thankyou

Expected Outcome:
Just want edge impulse to make the segregation of all animals properly.

Actual Outcome:
Not segregating the animals in proper manner, which is leading to improper detection.

Environment:

  • OS Version: Windows 10
#2

Hi @InfiAI

The issue is that you have a very imbalanced dataset, also the images are not the same in lighting and distance.

image
image1058×402 73.1 KB

Start with a smaller dataset that is balanced and similar lighting and distance then build upwards.

Best

Eoin

#3

Hi @Eoin,
Thankyou for the guidance, I also tried with samller dataset but its not working. I am not getting why? As the object in the image are nowhere similer to each other. And the reason of selecting a little diverse dataset is so that the inference must be able to detect in any condition, day, night, etc. I also made the dataset a bit balanced this time, but still the same.

Let me know what else can be done. I will again try it with only 2 objects.

Thank you.