cyberorigin/cyber_pipette_v0.1 · Datasets at Hugging Face

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	num_episodes int64 355 1.24k	task stringclasses 1 value
uint32	16	16	16	262,144	30	data/magvit2.ckpt	272,926	512	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	323,861	538	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	243,571	529	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	360,770	715	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	359,038	604	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	291,702	558	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	333,751	1,232	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	397,589	1,237	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	325,630	1,066	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	338,587	1,243	pipette
uint32	16	16	16	262,144	30	data/magvit2.ckpt	104,622	355	pipette

token_dtype

stringclasses

1 value

int64

vocab_size

int64

262k

int64

tokenizer_ckpt

stringclasses

1 value

num_images

int64

105k

398k

num_episodes

int64

355

1.24k

task

stringclasses

1 value

uint32

262,144

data/magvit2.ckpt

272,926

512

pipette

uint32

262,144

data/magvit2.ckpt

323,861

538

pipette

uint32

262,144

data/magvit2.ckpt

243,571

529

pipette

uint32

262,144

data/magvit2.ckpt

360,770

715

pipette

uint32

262,144

data/magvit2.ckpt

359,038

604

pipette

uint32

262,144

data/magvit2.ckpt

291,702

558

pipette

uint32

262,144

data/magvit2.ckpt

333,751

1,232

pipette

uint32

262,144

data/magvit2.ckpt

397,589

1,237

pipette

uint32

262,144

data/magvit2.ckpt

325,630

1,066

pipette

uint32

262,144

data/magvit2.ckpt

338,587

1,243

pipette

uint32

262,144

data/magvit2.ckpt

104,622

355

pipette

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 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": 10384598,
    "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"
}