stream.py

from streamlit_webrtc import webrtc_streamer
import numpy as np
import streamlit as st

import numpy as np
import av
import threading

import multiprocessing
from typing import List, Optional, Tuple

from pandas import DataFrame

import numpy as np
import pandas as pd
import streamlit as st
import torch
from torch import Tensor
from transformers import AutoFeatureExtractor, TimesformerForVideoClassification


from utils.frame_rate import FrameRate

np.random.seed(0)

st.set_page_config(
    page_title="TimeSFormer",
    page_icon="🧊",
    layout="wide",
    initial_sidebar_state="expanded",
    menu_items={
        "Get Help": "https://www.extremelycoolapp.com/help",
        "Report a bug": "https://www.extremelycoolapp.com/bug",
        "About": "# This is a header. This is an *extremely* cool app!",
    },
)


@st.cache_resource
# @st.experimental_singleton
def load_model(model_name: str):
    if "base-finetuned-k400" in model_name or "base-finetuned-k600" in model_name:
        feature_extractor = AutoFeatureExtractor.from_pretrained(
            "MCG-NJU/videomae-base-finetuned-kinetics"
        )
    else:
        feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
    model = TimesformerForVideoClassification.from_pretrained(model_name)
    return feature_extractor, model


lock = threading.Lock()

rtc_configuration = {
    "iceServers": [
        {
            "urls": "turn:relay1.expressturn.com:3478",
            "username": "efBRTY571ATWBRMP36",
            "credential": "pGcX1BPH5fMmZJc5",
        },
        # {
        #     "urls": [
        #         "stun:stun1.l.google.com:19302",
        #         "stun:stun2.l.google.com:19302",
        #         "stun:stun3.l.google.com:19302",
        #         "stun:stun4.l.google.com:19302",
        #     ]
        # },
    ],
}


def inference():
    if not img_container.ready:
        return

    inputs = feature_extractor(list(img_container.imgs), return_tensors="pt")

    with torch.no_grad():
        outputs = model(**inputs)
        logits: Tensor = outputs.logits

    # model predicts one of the 400 Kinetics-400 classes
    max_index = logits.argmax(-1).item()
    predicted_label = model.config.id2label[max_index]
    
    img_container.frame_rate.label = f"{predicted_label}_{logits[0][max_index]:.2f}%"

    TOP_K = 12
    # logits = np.squeeze(logits)
    logits = logits.squeeze().numpy()
    indices = np.argsort(logits)[::-1][:TOP_K]
    values = logits[indices]

    results: List[Tuple[str, float]] = []
    for index, value in zip(indices, values):
        predicted_label = model.config.id2label[index]
        # print(f"Label: {predicted_label} - {value:.2f}%")
        results.append((predicted_label, value))

    img_container.rs = pd.DataFrame(results, columns=("Label", "Confidence"))


class ImgContainer:
    def __init__(self, frames_per_video: int = 8) -> None:
        self.img: Optional[np.ndarray] = None  # raw image
        self.frame_rate: FrameRate = FrameRate()
        self.imgs: List[np.ndarray] = []
        self.frame_rate.reset()
        self.frames_per_video = frames_per_video
        self.rs: Optional[DataFrame] = None

    def add_frame(self, frame: np.ndarray):
        if len(img_container.imgs) >= frames_per_video:
            self.imgs.pop(0)
        self.imgs.append(frame)

    @property
    def ready(self):
        return len(img_container.imgs) == self.frames_per_video


def video_frame_callback(frame: av.VideoFrame) -> av.VideoFrame:
    img = frame.to_ndarray(format="bgr24")
    with lock:
        img_container.img = img
        img_container.frame_rate.count()
        img_container.add_frame(img)
        inference()
        img = img_container.frame_rate.show_fps(img)

    return av.VideoFrame.from_ndarray(img, format="bgr24")


def get_frames_per_video(model_name: str) -> int:
    if "base-finetuned" in model_name:
        return 8
    elif "hr-finetuned" in model_name:
        return 16
    else:
        return 96


st.title("TimeSFormer")

with st.expander("INTRODUCTION"):
    st.text(
        f"""Streamlit demo for TimeSFormer. 
        Number of CPU(s): {multiprocessing.cpu_count()}
    """
    )

model_name = st.selectbox(
    "model_name",
    (
        "facebook/timesformer-base-finetuned-k400",
        "facebook/timesformer-base-finetuned-k600",
        "facebook/timesformer-base-finetuned-ssv2",
        "facebook/timesformer-hr-finetuned-k600",
        "facebook/timesformer-hr-finetuned-k400",
        "facebook/timesformer-hr-finetuned-ssv2",
        "fcakyon/timesformer-large-finetuned-k400",
        "fcakyon/timesformer-large-finetuned-k600",
    ),
)
feature_extractor, model = load_model(model_name)


frames_per_video = get_frames_per_video(model_name)
st.info(f"Frames per video: {frames_per_video}")

img_container = ImgContainer(frames_per_video)

ctx = st.session_state.ctx = webrtc_streamer(
    key="snapshot",
    video_frame_callback=video_frame_callback,
    rtc_configuration=rtc_configuration,
)

if img_container.rs is not None:
    st.dataframe(img_container.rs)