Purple masks in output composite?

[raeesahhy]

Hi there,

Thanks for sharing this tool, it's very handy. I've tried it with 3D labels of my data and I'm getting some purple masks as you can see in the image below as part of the output. I'm not sure what these could mean. Sometimes throughout the stack the purple masks overlap with the deep blue FP masks but not always.

Also, I was wondering if there is a way to output the yellow masks from this composite in Fiji?

[Multimodal-Imaging-Center]

Hi,

The documentation is up-to-date for the 2D version but not for the 3D plugin. Below are the color significances in the 3D plugin:

• Yellow (TP): Objects common to both masks (GT and test) and validated by the given IoU threshold.
• Red (TP): True positives where objects are overestimated (larger in the test mask).
• Green (TP): True positives where objects are underestimated (smaller in the test mask).
• Orange (FN): False negatives where objects in the GT mask are merged into a single object in the test mask, with one object validated by the IoU threshold. The validated object is yellow, while others are orange.
• Cyan (FP): False positives where an object in the GT mask is split into multiple objects in the test mask. One object in the test mask will be yellow or cyan (depending on IoU validation), while others will be cyan.
• Blue (FP and FN): Objects common to both masks but not validated by the IoU threshold. The blue color covers the common area.
• Purple (FP): Objects common to both masks but not validated by the IoU threshold. The purple color covers the diverging area.
• Dark Red (FP): False positives where test mask objects do not match any object in the GT mask.
• Dark Green (FN): False negatives where GT mask objects do not match any object in the test mask.
• Grey: Objects eliminated based on the distance to the border parameter.
• Black: Pixels belonging to no object.

The overlapping of blue and purple is normal as it corresponds to the same objects. Extracting the yellow part is straightforward since the colors are represented by a Look-Up Table (LUT). The resulting image is coded in 8-bits, with yellow corresponding to the value 1. The color/value correspondence is as follows:

1. Yellow
2. Red
3. Green
4. Orange
5. Cyan
6. Blue
7. Purple
8. Dark Red
9. Dark Green
10. Grey

To extract the yellow part, you can use a threshold with both the upper and lower bounds set to 1.