AV Dataset Exploration

An Instance Segmentation Project using wandb as tracker and implementing inspections using the table features.

Created on February 19|Last edited on February 20

Comment

This is the data that are uploaded to wandb and are tracked for inspection.
﻿
project("gustiwinata", "mlops-course-001").artifact("bdd_simple_1k").membershipForAlias("v0").artifactVersion.file("eda_table.table.json")
 - 11 of 1000
525f8dcd-5b77b64e.jpg
None
1
1
0
1
0
1
0
00e9be89-00001840.jpg
None
1
1
1
1
0
1
0
47924e4b-6bef7091.jpg
None
1
1
1
1
1
1
0
325bf3b3-d7d0e527.jpg
None
1
1
1
1
1
1
0
File_Name
Images
Split
background
road
traffic light
traffic sign
person
vehicle
bicycle
1
2
3
4
5
6
7
8
9
10
11
The Data currently are imbalanced since the P1 and P2 value are out of balance. Most classes are well represented in the dataset except for bicycle which appear to be rare
﻿
﻿
project("gustiwinata", "mlops-course-001").artifact("bdd_simple_1k").membershipForAlias("v0").artifactVersion.file("eda_table.table.json")
 - 2 of 1000
File_Name
Split
background
road
traffic light
traffic sign
person
vehicle
bicycle
1
2
Images
This is the run for the training and inspection for the datasets.
﻿
runs.summary["pred_table"]
 - 80 of 100
Image
background IoU
road IoU
traffic light IoU
traffic sign IoU
person IoU
vehicle IoU
bicycle IoU
77
78
79
80
Run: devoted-sunset-81
﻿
These are the charts that are tracked with wandb, feast your eyes
﻿
Run: devoted-sunset-81
﻿
﻿

Add a comment