demo.py

import os
import shutil
import time
import numpy as np
import torch
import torchvision.transforms as T
from decord import VideoReader, cpu
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

# model setting
model_path = './'
device = torch.device("cuda:0")
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModel.from_pretrained(model_path, trust_remote_code=True).half().to(device).to(torch.bfloat16)

IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([T.Lambda(lambda img: img.convert("RGB") if img.mode != "RGB" else img), T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC), T.ToTensor(), T.Normalize(mean=MEAN, std=STD)])
    return transform


def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float("inf")
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio


def dynamic_preprocess(image, min_num=1, max_num=6, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set((i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = ((i % (target_width // image_size)) * image_size, (i // (target_width // image_size)) * image_size, ((i % (target_width // image_size)) + 1) * image_size, ((i // (target_width // image_size)) + 1) * image_size)
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images


def load_image(image, input_size=448, max_num=6):
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values


def get_index(bound, fps, max_frame, first_idx=0, num_segments=32):
    if bound:
        start, end = bound[0], bound[1]
    else:
        start, end = -100000, 100000
    start_idx = max(first_idx, round(start * fps))
    end_idx = min(round(end * fps), max_frame)
    seg_size = float(end_idx - start_idx) / num_segments
    frame_indices = np.array([int(start_idx + (seg_size / 2) + np.round(seg_size * idx)) for idx in range(num_segments)])
    return frame_indices

def get_num_frames_by_duration(duration):
        local_num_frames = 1  
        num_segments = int(duration // local_num_frames)
        if num_segments == 0:
            num_frames = local_num_frames
        else:
            num_frames = local_num_frames * num_segments
        
        num_frames = min(512, num_frames)
        num_frames = max(1, num_frames)
        
        return num_frames

def load_video(video_path, bound=None, input_size=448, max_num=1, num_segments=32, get_frame_by_duration = False):
    vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
    max_frame = len(vr) - 1
    fps = float(vr.get_avg_fps())
    video_name = os.path.splitext(os.path.basename(video_path))[0]
    save_dir = f'./examples/frames/{video_name}'
    if os.path.exists(save_dir):
        save_flag = False
    else:
        save_flag = True
        os.makedirs(save_dir, exist_ok=True)
        destination_path = f'./examples/videos/{os.path.basename(video_path)}'
        os.makedirs(destination_path, exist_ok=True)
        shutil.copy(video_path, destination_path)
        print(f"Video copied to {destination_path}")
    pixel_values_list, num_patches_list = [], []
    transform = build_transform(input_size=input_size)
    if get_frame_by_duration:
        duration = max_frame / fps
        num_segments = get_num_frames_by_duration(duration)
    frame_indices = get_index(bound, fps, max_frame, first_idx=0, num_segments=num_segments)
    for i in range(len(frame_indices)):
        img = Image.fromarray(vr[frame_indices[i]].asnumpy()).convert("RGB")
        if save_flag:
            save_path = os.path.join(save_dir, f'frame_{i+1}.png')
            img.save(save_path)
        img = dynamic_preprocess(img, image_size=input_size, use_thumbnail=True, max_num=max_num)
        pixel_values = [transform(tile) for tile in img]
        pixel_values = torch.stack(pixel_values)
        num_patches_list.append(pixel_values.shape[0])
        pixel_values_list.append(pixel_values)
    pixel_values = torch.cat(pixel_values_list)
    return pixel_values, num_patches_list

# evaluation setting
max_num_frames = 512
generation_config = dict(
    max_new_tokens=1024,
    num_beams=1,
    repetition_penalty = 1.05
)
video_path = "./demo.mp4"

temporal_questions = {
    2: "Where is the man lying in the video?",
    4: "What could be the possible relationship between him and the person next to him?",
    8: "What is the woman in the video doing?",
    10: "What is the expression of the woman in the video?",
    12: "What is his reaction?",
    13: "Is there a thermos on the table beside the hospital bed?",
    14: "Is there any tissue on the table?",
    20: "What color is the woman's clothing?",
    26: "How does the color of the bed differ from it?",
    41: "What is the girl in the video doing?",
    46: "What does the boy in the video say?",
    49: "How is his tone when he speaks?",
    50: "From what he said, could this woman be his mother?",
    64: "What is the expression in the boy's eyes?",
    73: "What else does the boy say?",
    74: "What animal is this toy?",
    75: "What color is the toy in the boy's memory?",
    78: "What's the difference between the scene with the door and the scene with the frog toy that appeared before?",
    79: "Is there a lock on the door?",
    81: "Are there any plants on the hospital room window?",
    87: "What's on the boy's back?",
    88: "What did the girl do to the scar?",
    92: "Does the girl have any special facial expression while wiping the scar?"
}

with torch.no_grad():
    pixel_values, num_patches_list = load_video(video_path, max_num=1, get_frame_by_duration=True)
    pixel_values = pixel_values.to(torch.bfloat16).to(model.device)
    batch_frame = 1
    start_time = time.time()
    chat_history = None
    question = ''
    
    for i in range(0, 100, batch_frame):
        video_frame = "".join([f"Frame-{i+j+1}: <image>\n" for j in range(batch_frame)])
        question += video_frame
        
        if (i + 1) in temporal_questions:
            question += temporal_questions[i + 1] + "\n"
            output_last, chat_history = model.chat(
                tokenizer, 
                pixel_values[:i+batch_frame, ...], 
                question, 
                generation_config, 
                num_patches_list=num_patches_list[:i+batch_frame], 
                history=chat_history, 
                return_history=True
            )
            print(f"{'Frame' + str(i+1):<15} {'Q: ' + temporal_questions[i+1]:<50}")
            print(f"{'':<15} {'A: ' + output_last:<50}")
            question = ''
        else:
            print(f"{'Frame' + str(i+1):<15} {'Keep watching...':<50}")
                
end_time = time.time()
print("Program runtime:", end_time - start_time, "seconds")