I want to return frames to the client instead of showing them in the browser.

[mle-usti]

Hey, I am working on a realtime object detection API, I really like your work, but I need the frames from flask to be returned to the client, whether the client is using an iPhone, an Android or just a python script as client.

[codectl]

Running both processes in the same machine is not an option ?

[mle-usti]

Nope, I need the stream from the client's machine to be sent to my server and then processed there and returned back to the client.

[codectl]

Your client is the process reading the frames from the camera so technically the client has direct access to those frames . If that is not the setup you looking for perhaps lay out some details on what you trying to achieve to see how to go about that.

[mle-usti]

I am working on an object detection API in Flask and I want to get a real-time video stream from the client - which might be from an Android app, an iOS app, or just from a python script running on Windows/Linux - so that I can detect objects in real-time with the detection model (yolov4) on the remote server with Flask. I know it can be converted to tflite but I don't want to use that.

After some work: I was able to do this in python with Server.py running on a remote machine:

Client.py

def RT_Rec():
    camera = cv2.VideoCapture(0)
    try:
        while True:
            ret, frame = camera.read()
            content_type = 'image/jpeg'
            headers = {'content-type': content_type}
            url = "https://[REMOTEMACHINEIP]/RealTime"
            _, img_encoded = cv2.imencode('.jpg', frame)       
            response = requests.post(url, data=img_encoded.tobytes(), headers= headers)
            if b"[SUCCESS]" in response.content or b"[ERROR]" in response.content :
                print(response.content.decode('utf-8'))
                break
            else:
                nparr = np.frombuffer(response.content,np.uint8)
                img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
                cv2.imshow("camera", img)
            if cv2.waitKey(1) & 0xFF == ord("q"):
                break
    except Exception as e:
        print(e)
    finally:
        camera.release()
        cv2.destroyAllWindows()

Server.py

@endpoints.route("/RealTime",methods=["GET","POST"])
def realTime():
    if request.method == "POST":
        img = request.data
        nparr = np.fromstring(img,np.uint8)
        img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
        det,outImg = processFrame(img)
        if det:
            msg = "[SUCCESS] Processed"
            return msg
        _, img_encoded = cv2.imencode('.jpg', outImg)
        return img_encoded.tobytes()
    else:
        return "[ERROR] Invalid Request", 500

Now I what I am looking for is to implement the same thing for an Android/iOS app

[umairahmadh]

@mle-usti I am interested in your work. I tried to replicate what you did there by hosting the server script at pythonanywhere.com and having the client in my PC. But I am facing some issues getting it working. Can you plese help me there?

[yaswanthsai002]

@mle-usti I am also feeling the same which is mentioned above. Could you please solve the issue and give us a solution?