token_dtype
stringclasses 1
value | s
int64 16
16
| h
int64 16
16
| w
int64 16
16
| vocab_size
int64 262k
262k
| hz
int64 15
15
| tokenizer_ckpt
stringclasses 1
value | num_images
int64 19.3k
70.9k
|
---|---|---|---|---|---|---|---|
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 45,217 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 35,949 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 51,201 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 70,944 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 54,056 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 54,632 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 38,277 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 19,252 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 32,772 |
uint32 | 16 | 16 | 16 | 262,144 | 15 | imagenet_256_L.ckpt | 28,951 |
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:
alita # dataset root path
└── data/
├── metadata_Traj01.json
├── segment_ids_Traj01.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_Traj01.bin # 16x16 image patches at 15hz, 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": "ALITA",
"total_episodes": ,
"total_frames": ,
"token_dtype": "uint32",
"vocab_size": 262144,
"fps": 15,
"language_annotation": "None",
}
- Downloads last month
- 10