File size: 7,039 Bytes
0710b5c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
"""
step2_prepare_data.py
======================
Task 4 β€” Component 2: Load COCO validation data for diversity analysis
and build style-labelled caption sets for steering vector extraction.

Two public APIs
---------------
1. load_val_data(processor, n=200, batch_size=4) -> DataLoader
   Loads the first ``n`` MS-COCO 2017 validation images.
   Each batch yields {"pixel_values": Tensor, "captions": list[str], "image_ids": list[int]}.

2. build_style_sets(n=500) -> dict[str, list[str]]
   Loads COCO validation captions and partitions them by word-count length:
       short    : ≀ 8 words
       medium   : 9–14 words
       detailed : β‰₯ 15 words
   Returns {"short": [...], "medium": [...], "detailed": [...]}

Standalone usage
----------------
    export PYTHONPATH=.
    venv/bin/python task/task_04/step2_prepare_data.py
"""

import os
import sys
import torch
from torch.utils.data import Dataset, DataLoader

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))


# ─────────────────────────────────────────────────────────────────────────────
# COCO Diversity Dataset
# ─────────────────────────────────────────────────────────────────────────────

class COCODiversityDataset(Dataset):
    """Wraps COCO validation images for diversity analysis.

    Each item returns:
        pixel_values : (3, H, W) float tensor (processor-normalised)
        caption      : str  β€” first reference caption from COCO
        image_id     : int
    """

    def __init__(self, hf_dataset, processor, max_n: int = 200):
        self.data      = hf_dataset.select(range(min(max_n, len(hf_dataset))))
        self.processor = processor

    def __len__(self):
        return len(self.data)

    def __getitem__(self, idx):
        item   = self.data[idx]
        image  = item["image"].convert("RGB")
        inputs = self.processor(images=image, return_tensors="pt")
        pixel_values = inputs["pixel_values"].squeeze(0)

        # First COCO reference caption
        captions = item.get("captions", item.get("sentences_raw", [""]))
        caption  = captions[0] if isinstance(captions, list) else captions

        return {
            "pixel_values": pixel_values,
            "caption": caption,
            "image_id": idx,
            "image": image,
        }


def _collate(batch):
    pixel_values = torch.stack([b["pixel_values"] for b in batch])
    captions     = [b["caption"]  for b in batch]
    image_ids    = [b["image_id"] for b in batch]
    images       = [b["image"]    for b in batch]
    return {
        "pixel_values": pixel_values, 
        "captions": captions, 
        "image_ids": image_ids,
        "images": images
    }


def load_val_data(processor, n: int = 200, batch_size: int = 4):
    """
    Load the first ``n`` COCO 2017 validation images as a DataLoader.

    Args:
        processor  : BlipProcessor (from step1_load_model)
        n          : number of images to load (default 200)
        batch_size : images per batch (default 4)

    Returns:
        torch.utils.data.DataLoader
    """
    from datasets import load_dataset
    print("=" * 62)
    print("  Task 4 β€” Step 2: Prepare COCO Validation Data")
    print("=" * 62)
    print(f"  Loading COCO 2017 validation split (first {n} images)…")

    hf_ds = load_dataset("whyen-wang/coco_captions", split="validation", trust_remote_code=True)
    print(f"  βœ… COCO val split loaded ({len(hf_ds):,} total images)")

    dataset = COCODiversityDataset(hf_ds, processor, max_n=n)
    loader  = DataLoader(
        dataset,
        batch_size=batch_size,
        shuffle=False,
        collate_fn=_collate,
        num_workers=0,
    )

    print(f"  Dataset size : {len(dataset)} images")
    print(f"  Batch size   : {batch_size}")
    print(f"  Num batches  : {len(loader)}")
    print("=" * 62)
    return loader


# ─────────────────────────────────────────────────────────────────────────────
# Style caption sets (for steering vector extraction)
# ─────────────────────────────────────────────────────────────────────────────

def build_style_sets(n: int = 500) -> dict:
    """
    Partition the first ``n`` COCO validation captions into three style buckets
    based on word count:

        short    : ≀ 8 words
        medium   : 9–14 words
        detailed : β‰₯ 15 words

    Args:
        n: Maximum number of COCO captions to scan

    Returns:
        dict with keys 'short', 'medium', 'detailed', each a list[str].
    """
    from datasets import load_dataset
    print("  Building style caption sets from COCO val …")

    hf_ds  = load_dataset("whyen-wang/coco_captions", split="validation", trust_remote_code=True)
    short, medium, detailed = [], [], []

    for item in hf_ds.select(range(min(n, len(hf_ds)))):
        captions = item.get("captions", item.get("sentences_raw", []))
        if isinstance(captions, str):
            captions = [captions]
        for cap in captions:
            wc = len(cap.split())
            if wc <= 8:
                short.append(cap)
            elif wc <= 14:
                medium.append(cap)
            else:
                detailed.append(cap)

    print(f"  Style sets: short={len(short)}, medium={len(medium)}, detailed={len(detailed)}")
    return {"short": short, "medium": medium, "detailed": detailed}


# ─────────────────────────────────────────────────────────────────────────────
# Standalone entrypoint
# ─────────────────────────────────────────────────────────────────────────────

if __name__ == "__main__":
    from step1_load_model import load_model
    model, processor, device = load_model()

    # Test data loader
    loader = load_val_data(processor, n=20, batch_size=4)
    batch  = next(iter(loader))
    print(f"\n  Sample batch β€” pixel_values shape : {batch['pixel_values'].shape}")
    print(f"  Sample captions : {batch['captions'][:2]}")

    # Test style sets
    sets = build_style_sets(n=100)
    for style, caps in sets.items():
        sample = caps[0] if caps else "(none)"
        print(f"  {style:8s} ({len(caps):3d} caps)  e.g. '{sample[:60]}'")