services/extractor.py

import cv2
import torch
import numpy as np
from PIL import Image
from torchvision import models, transforms
from config import DEVICE, FRAME_RATE
from tqdm import tqdm
from services.model_loader import batch_inference

# Load GoogLeNet once
from torchvision.models import GoogLeNet_Weights
weights = GoogLeNet_Weights.DEFAULT
googlenet = models.googlenet(weights=weights).to(DEVICE).eval()

feature_extractor = torch.nn.Sequential(
    googlenet.conv1,
    googlenet.maxpool1,
    googlenet.conv2,
    googlenet.conv3,
    googlenet.maxpool2,
    googlenet.inception3a,
    googlenet.inception3b,
    googlenet.maxpool3,
    googlenet.inception4a,
    googlenet.inception4b,
    googlenet.inception4c,
    googlenet.inception4d,
    googlenet.inception4e,
    googlenet.maxpool4,
    googlenet.inception5a,
    googlenet.inception5b,
    googlenet.avgpool,
    torch.nn.Flatten()
)
feature_extractor = feature_extractor.eval()

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

def extract_features(video_path):
    cap = cv2.VideoCapture(video_path)
    frames = []
    indices = []
    total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    # total_frames = 300 # TEMP
    print(f"Total frames in video: {total_frames}")

    for idx in tqdm(range(total_frames)):
        cap.set(cv2.CAP_PROP_POS_FRAMES, idx)
        ret, frame = cap.read()
        if not ret:
            break

        # process frame
        frame = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
        frame = transform(frame)

        frames.append(frame)
        indices.append(idx)

    cap.release()

    frames = torch.stack(frames).to(DEVICE)
    print("Features before GoogleNet extraction:", frames.shape)
    frames = batch_inference(model=feature_extractor, input=frames, batch_size=32)
    print("Features after GoogleNet extraction:", frames.shape)

    return frames, indices