token_dtype
stringclasses
1 value
s
int64
16
16
h
int64
16
16
w
int64
16
16
vocab_size
int64
262k
262k
hz
int64
30
30
tokenizer_ckpt
stringclasses
1 value
num_images
int64
105k
398k
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
272,926
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
323,861
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
243,571
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
360,770
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
359,038
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
291,702
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
333,751
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
397,589
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
325,630
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
338,587
uint32
16
16
16
262,144
30
imagenet_256_L.ckpt
104,622

CyberOrigin Dataset

Our data includes information from home services, the logistics industry, and laboratory scenarios. For more details, please refer to our Offical Data Website

contents of the dataset:

cyber_pipette # dataset root path
  └── data/
      ├── metadata_ID1_240808.json
      ├── segment_ids_ID1_240808.bin # for each frame segment_ids uniquely points to the segment index that frame i came from. You may want to use this to separate non-contiguous frames from different videos (transitions).
      ├── videos_ID1_240808.bin # 16x16 image patches at 30hz, each patch is vector-quantized into 2^18 possible integer values. These can be decoded into 256x256 RGB images using the provided magvit2.ckpt weights.
      ├── ...
  └── ...
{
    "task": "Pipette",
    "total_episodes": 8589,
    "total_frames": 3352047,
    "token_dtype": "uint32",
    "vocab_size": 262144,
    "fps": 30,
    "manipulation_type": "Bi-Manual",
    "language_annotation": "None",
    "scene_type": "Table Top",
    "data_collect_method": "Directly Collection on Human"
}
Downloads last month
157
Edit dataset card