Download

You can download the dataset from the following link. You can download each object datset individually.

Format

Inside each object dataset folder you will find everything except the test dataset that is kept private.

The json files have the same format as the original NeRF synthetic dataset.

Name	Description
`train_images`	The training dataset containing 1000 images
`val_images`	The validation dataset containing 500 images
`train_all.json`	The camera info of the training dataset for the split Train 1000
`train_all_500.json`	The camera info of the training dataset for the split Train 500
`train_all_250.json`	The camera info of the training dataset for the split Train 250
`train_all_100.json`	The camera info of the training dataset for the split Train 100
`val_all.json`	The camera info of the validation dataset for the splits Train 1000 / Train 500 / Train 250 / Train 100
`test_all.json`	The camera info of the test dataset for the splits Train 1000 / Train 500 / Train 250 / Train 100
`train_0.json`	The camera info of the training dataset for the sub-split Train 0
`train_1.json`	The camera info of the training dataset for the sub-split Train 1
`train_2.json`	The camera info of the training dataset for the sub-split Train 2
`train_3.json`	The camera info of the training dataset for the sub-split Train 3
`train_4.json`	The camera info of the training dataset for the sub-split Train 4
`train_5.json`	The camera info of the training dataset for the sub-split Train 5
`train_6.json`	The camera info of the training dataset for the sub-split Train 6
`train_7.json`	The camera info of the training dataset for the sub-split Train 7
`val_0.json`	The camera info of the validation dataset for the sub-split Train 0
`val_1.json`	The camera info of the validation dataset for the sub-split Train 1
`val_2.json`	The camera info of the validation dataset for the sub-split Train 2
`val_3.json`	The camera info of the validation dataset for the sub-split Train 3
`val_4.json`	The camera info of the validation dataset for the sub-split Train 4
`val_5.json`	The camera info of the validation dataset for the sub-split Train 5
`val_6.json`	The camera info of the validation dataset for the sub-split Train 6
`val_7.json`	The camera info of the validation dataset for the sub-split Train 7
`test_0.json`	The camera info of the test dataset for the sub-split Train 0
`test_1.json`	The camera info of the test dataset for the sub-split Train 1
`test_2.json`	The camera info of the test dataset for the sub-split Train 2
`test_3.json`	The camera info of the test dataset for the sub-split Train 3
`test_4.json`	The camera info of the test dataset for the sub-split Train 4
`test_5.json`	The camera info of the test dataset for the sub-split Train 5
`test_6.json`	The camera info of the test dataset for the sub-split Train 6
`test_7.json`	The camera info of the test dataset for the sub-split Train 7

Training instructions

You can submit your results for two categories:

evenly distributed splits
densely localized sub-splits

EVENLY DISTRIBUTED SPLITS

For each object dataset, you have to train on the splits: train_all.json, train_all_500.json, train_all_250.json, train_all_100.json. For each split you have to validate and test on val_all.jsonand test_all.json.

DENSELY LOCALIZED SUB-SPLITS

For each object dataset, you have to train on all the 8 sub-splits train_*.json. For each sub-split you have to validate and test on all the sub-splits val_*.json and test_*.json.