Spaces:
Sleeping
Sleeping
File size: 2,536 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 | """
step2_encode_image.py
======================
STEP 2 — Encode an image through BLIP's Vision Transformer (ViT).
Responsibilities:
- Accept a PIL image.
- Run it through the ViT image encoder.
- Return encoder_hidden_states (197 patch tokens × 768 dim).
- Also return the encoder_attention_mask.
This is kept separate so the expensive ViT encode is run ONCE for
however many words we later generate — zero redundant computation.
Shape of the output:
encoder_hidden_states : (1, 197, 768)
encoder_mask : (1, 197) — all-ones (no padding)
"""
import os
import sys
import torch
from PIL import Image
_THIS_DIR = os.path.dirname(os.path.abspath(__file__))
_PROJECT_ROOT = os.path.dirname(os.path.dirname(_THIS_DIR))
if _PROJECT_ROOT not in sys.path:
sys.path.insert(0, _PROJECT_ROOT)
# ────────────────────────────────────────────────────────────────────────────
_ENCODE_SIZE = 224 # ViT expects 224×224 → 14×14 patches of size 16
def encode_image(model, processor, device, image_pil: Image.Image, verbose: bool = True):
"""
Encode a PIL image through BLIP's ViT backbone.
Args:
model – BlipForConditionalGeneration.
processor – BlipProcessor.
device – torch.device.
image_pil – Any PIL image (will be resized to 224×224 internally).
verbose – Print progress.
Returns:
image_224 – PIL image resized to 224×224.
encoder_hidden – Tensor (1, 197, 768), detached, no grad.
encoder_mask – Tensor (1, 197), all-ones.
"""
image_224 = image_pil.resize((_ENCODE_SIZE, _ENCODE_SIZE), Image.LANCZOS)
inputs = processor(images=image_224, return_tensors="pt").to(device)
if verbose:
print(f"📷 Image resized to {_ENCODE_SIZE}×{_ENCODE_SIZE} and encoded through ViT …")
with torch.no_grad():
vision_out = model.vision_model(pixel_values=inputs["pixel_values"])
# Shape: (1, 197, 768) — 1 [CLS] token + 196 patch tokens
encoder_hidden = vision_out[0].detach().requires_grad_(False)
encoder_mask = torch.ones(encoder_hidden.size()[:-1], dtype=torch.long, device=device)
if verbose:
print(f"✅ Encoder output shape: {encoder_hidden.shape} "
f"(1 CLS + {encoder_hidden.shape[1]-1} patches)")
return image_224, encoder_hidden, encoder_mask
|