Dataset Viewer
Auto-converted to Parquet
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
51k
1.35M
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,232,975
uint32
16
16
16
262,144
30
data/magvit2.ckpt
916,648
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,085,940
uint32
16
16
16
262,144
30
data/magvit2.ckpt
829,177
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,217,781
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,212,761
uint32
16
16
16
262,144
30
data/magvit2.ckpt
51,038
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,150,321
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,352,994
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,106,470
uint32
16
16
16
262,144
30
data/magvit2.ckpt
1,154,239
uint32
16
16
16
262,144
30
data/magvit2.ckpt
358,267

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 train/val data:

train # train dataset root folder
  └── ID1/
    └── 2024-08-07/
        ├── metadata.json
        ├── segment_ids.bin # for each frame segment_ids[i] 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.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.
  └── ID2/
    └── 2024-08-08/
        ├── ...

val # val dataset root folder
  └── ID1/
    └── 2024-08-05/
        ├── metadata.json
        ├── segment_ids.bin
        ├── videos.bin
  └── ID2/
    └── 2024-08-06/
        ├── ...
Downloads last month
44