| |
| """Debug script to inspect pkl structure""" |
| import pickle |
| import numpy as np |
|
|
| pkl_file = 'results/declip_vitb16/predictions.pkl' |
|
|
| print("Loading pkl...") |
| with open(pkl_file, 'rb') as f: |
| results = pickle.load(f) |
|
|
| print(f"Type: {type(results)}") |
| print(f"Length: {len(results)}") |
|
|
| |
| for i in range(min(3, len(results))): |
| result = results[i] |
| print(f"\n=== Result {i} ===") |
| print(f" Type: {type(result)}") |
| if isinstance(result, (list, tuple)): |
| print(f" Length: {len(result)}") |
| for j, item in enumerate(result[:3]): |
| print(f" [{j}] Type: {type(item)}, Shape: {getattr(item, 'shape', 'N/A')}, Len: {len(item) if hasattr(item, '__len__') else 'N/A'}") |
| if hasattr(item, '__len__') and len(item) > 0: |
| if hasattr(item, 'shape'): |
| print(f" First elem shape: {item[0].shape if len(item) > 0 else 'empty'}") |
| if len(item) > 0 and len(item[0]) > 0: |
| print(f" First box: {item[0]}") |
|
|
| |
| print("\n=== Finding non-empty predictions ===") |
| for i in range(min(100, len(results))): |
| result = results[i] |
| if isinstance(result, (list, tuple)): |
| for j, bboxes in enumerate(result): |
| if hasattr(bboxes, '__len__') and len(bboxes) > 0: |
| print(f"Image {i}, Class {j}: {len(bboxes)} boxes") |
| print(f" Type: {type(bboxes)}, Shape: {bboxes.shape if hasattr(bboxes, 'shape') else 'N/A'}") |
| if len(bboxes) > 0: |
| print(f" First box: {bboxes[0]}") |
| print(f" First box type: {type(bboxes[0])}") |
| print(f" First box shape: {bboxes[0].shape if hasattr(bboxes[0], 'shape') else len(bboxes[0])}") |
| break |
| else: |
| continue |
| break |
|
|