Possible overlay bug with binary segmentation masks

• Weights and Biases version: wandb, version 0.8.32

• Python version: Python 3.6.8

• Operating System: Linux

Description

Semantic Segmentation usecase for binary image masks feature for 3D volumes in grayscale

What I Did

Here’s the code to reproduce:

Malfunctioning code: The dashboard shows overlays only for class with label 0.

import numpy as np
import wandb


wandb.init(project='prototyping', entity='maastro-clinic')


target = np.random.randint(2, size=(32, 256, 256)).astype(np.uint8)
output = np.random.randint(2, size=(32, 256, 256)).astype(np.uint8)
data = np.random.randint(256, size=(32, 256, 256)).astype(np.uint8)


images = []

idx = 0
class_labels = {
    0: 'background',
    1: "nodule"
}

for im, t, o in zip(data, target, output):
    idx += 1        
    images.append(wandb.Image(im, masks={"predictions": {"mask_data": o, "class_labels": class_labels}, \
        "ground_truth": {"mask_data": t, "class_labels": class_labels}}, caption='CT'))


wandb.log({"Predictions": images})

Changing the masks to have values from 0 to 2 and adding a class label makes all the 3 overlays work fine.

import numpy as np
import wandb


wandb.init(project='prototyping', entity='maastro-clinic')


target = np.random.randint(3, size=(32, 256, 256)).astype(np.uint8)
output = np.random.randint(3, size=(32, 256, 256)).astype(np.uint8)
data = np.random.randint(256, size=(32, 256, 256)).astype(np.uint8)


images = []

idx = 0
class_labels = {
    0: 'background',
    1: "nodule",
    2: "nodule2"

}

for im, t, o in zip(data, target, output):
    idx += 1        
    images.append(wandb.Image(im, masks={"predictions": {"mask_data": o, "class_labels": class_labels}, \
        "ground_truth": {"mask_data": t, "class_labels": class_labels}}, caption='CT'))


wandb.log({"Predictions": images})

Am I missing something here?

[nbardy]

Hey @surajpaib thanks for reporting these bugs. I want to get both of these fixed for you.

I’ve got a PR up to fix the binary image masks issue #985. I’m looking at the other issue and I’ll get a fix out this week.

[nbardy]

@surajpaib I’m going to close this issue. Feel free to open a new ticket if you find the other issue causing problems.