sunmengxi/panorama-video-process

Panorama Video Process - Data and Outputs

This dataset contains the test data and outputs from three modes for memory tests.

Structure

• test_data/: 15 selected test samples from https://huggingface.co/datasets/quchenyuan/360x_dataset_LR/tree/main
• outputs/two-turn/: processed-video by two-turn mode + metadata + visualized final away duration distribution
• outputs/away-target-away-target same/: processed-video by away-target-away-target same mode + metadata + visualized final away duration distribution
• outputs/away-target-away-target opposite/: processed-video by away-target-away-target opposite mode + metadata + visualized final away duration distribution

Modes

1. two_turn: target-away-target

Temporal order:

target hold
  -> smooth turn to away
  -> away hold
  -> smooth turn back to target
  -> final target hold

The goal of this condition is: the target is seen from the very first frame, then the view leaves the target, and finally, within the prediction window, the view turns back to the same target from the away direction.

2. away_target_away_target + same

Temporal order:

entry away hold
  -> smooth turn to target
  -> middle target hold
  -> smooth turn to exit away
  -> exit away hold
  -> smooth turn back to target
  -> final target hold

same means the entry away and exit away use the same away yaw and away FOV.

3. away_target_away_target + opposite

The temporal order is the same as same：

entry away hold
  -> smooth turn to target
  -> middle target hold
  -> smooth turn to exit away
  -> exit away hold
  -> smooth turn back to target
  -> final target hold

opposite means the entry away and exit away lie in opposite directions relative to the target: if the entry away is the target yaw plus a positive delta, then the exit away uses a negative delta, and vice versa. The FOV sizes of the two away segments are identical, but their yaws differ.

Usage

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="sunmengxi/panorama-video-process",
    repo_type="dataset"
)
print(path)