chaenayo commited on
Commit
01320e8
·
verified ·
1 Parent(s): 867647d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -0
README.md CHANGED
@@ -59,6 +59,10 @@ The PODS dataset is new a benchmark for personalized vision tasks. It includes:
59
  * 71-201 test images per instance with classification label annotations.
60
  * 12 test images per instance (3 per split) with segmentation annotations.
61
 
 
 
 
 
62
  Metadata is stored in two files:
63
  * `pods_info.json`:
64
  * `classes`: A list of class names
@@ -70,6 +74,7 @@ Metadata is stored in two files:
70
  * `split`: One of `[train, test]` indicating if the image is in the train or test set for that class.
71
  * `test_split`: For images in the `test` split, denotes which distribution-shift test split the image is in: One of `[in_distribution, pose, distractors, pose_and_distractors]`
72
 
 
73
  ## Using PODS
74
 
75
  ### Loading the dataset using HuggingFace
 
59
  * 71-201 test images per instance with classification label annotations.
60
  * 12 test images per instance (3 per split) with segmentation annotations.
61
 
62
+ PODS is split *class-wise* into a validation set (6 classes per semantic category) and a test set (14 classes per semantic category). All test performance reported in our paper is from the test set of classes.
63
+
64
+ *Within each class*, images are divided into a train/retrieval set (3 images) and a test/query set. The test/query set is then further divided into 4 test splits reflecting different distribution shifts.
65
+
66
  Metadata is stored in two files:
67
  * `pods_info.json`:
68
  * `classes`: A list of class names
 
74
  * `split`: One of `[train, test]` indicating if the image is in the train or test set for that class.
75
  * `test_split`: For images in the `test` split, denotes which distribution-shift test split the image is in: One of `[in_distribution, pose, distractors, pose_and_distractors]`
76
 
77
+
78
  ## Using PODS
79
 
80
  ### Loading the dataset using HuggingFace