Memory leakage when doing ML inference

Using Streamlit
1 
import cv2
import supervision as sv

from collections import defaultdict, deque

from inference import InferencePipeline
from inference.core.interfaces.camera.entities import VideoFrame
 
import streamlit as st
from PIL import Image

frame_display = st.empty()

def my_custom_sink(predictions: dict, video_frame: VideoFrame):
    #Parts of the code related to the problem
    annotated_frame_rgb = cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB)
    image_pil = Image.fromarray(annotated_frame_rgb)
    frame_display.image(image_pil, caption="Annotated Frame", use_column_width=True)

pipeline1 = InferencePipeline.init(
    model_id= "yolov8n-640",
    video_reference= SOURCE_VIDEO_PATH,
    on_prediction=my_custom_sink,
)

pipeline1.start()
pipeline1.join()

My RAM usage kept rising every time the script ran, RAM usage only leveled out when I commented the frame_display.image() part. Are there any suggestions to avoid this problem? Thank you in advance

2

Hey @Ardhana_Putra ! Thanks for posting and Welcome to the community!

Looks like we are talking over in the GitHub issue, so I think that seems best to keep things there. Here it is for anyone following.

3

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.