evo-output-serialization

Serializes analysis results to the required output formats.

Key Functions

• dense_to_csr_manual(dense_mask) — Converts 2D bool array to CSR (data, indices, indptr)
• save_csr_masks_npz(mask_list, output_file, shape) — Saves masks to NPZ with CSR format
• export_motion_labels_json(motion_results, output_path) — Saves motion labels to JSON

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-output-serialization/scripts')
from utils import save_csr_masks_npz, export_motion_labels_json

# motion_results: list of (sample_idx_prev, sample_idx_curr, labels_list)
export_motion_labels_json(motion_results, '/root/pred_instructions.json')

# mask_list: list of boolean 2D arrays, shape: [H, W]
save_csr_masks_npz(mask_list, '/root/pred_dyn_masks.npz', shape=[H, W])

CSR Format Details

For each frame i, stores:

• f_{i}_data: bool array of True values
• f_{i}_indices: int32 column indices of nonzero elements
• f_{i}_indptr: int32 array of length H+1 (row pointers)
• shape: [H, W] as int32 array

JSON Format

{
    "0->1": ["Pan Right"],
    "1->2": ["Stay"],
    ...
}

Keys are "sample_idx_prev->sample_idx_curr" (0-based sample indices).