bug: After extending the yolo-flutter plugin, I found that passing a 91KB image byte array from Flet to Dart takes up to 60-100ms. This severely reduces the frame rate on the Flet side, limiting the maximum data exchange rate to just 10 frames per second.

[yst-one]

Duplicate Check

• I have searched the opened issues and there are no duplicates

Describe the bug

flutter Inference time

flet inference time

Code sample

Code

[Paste your code here]

To reproduce

1

Expected behavior

No response

Screenshots / Videos

Captures

[Upload media here]

Operating System

Windows

Operating system details

11

Flet version

0.85.3

Regression

Yes, it used to work in a previous Flet version (please specify the version in additional details)

Suggestions

No response

Logs

Logs

[Paste your logs here]

Additional details

No response

[ndonkoHenri]

Will be nice if you could comeup with the simplest possible repro, so we test it out on our end.

[yst-one]

提供一个最简单的复现步骤，以便我们进行测试，那就太好了。

https://github.com/yst-one/yolo_flet

[ndonkoHenri]

Steps/Code to repro the issue?

[yst-one]

    async def detect() :
        log.debug("/。。。。。。。。。。开始识别中。。。。。。。。。。")
        log.info(f"{find_template_location(target_img_path="./assets/test.jpg", template_img_path="./assets/sub.png", threshold=0.5)}")


        with open("./assets/test.jpg", "rb") as f:

            image_bytes = f.read()
            for i in range(10000):
                start = time.time()
                # a:list[types.DetectionResult] =await  yolo_flet.detect(image_bytes)

                await update_text(i)
                time.sleep(0.1)
                end = time.time()
                log.debug("耗时"+str((end - start) * 1000))
                # for i in a:
                #     log.debug(i)

    async def update_text(i: int):
        text.value = f"这是第{i}次推理"
        text.update()

Besides, after I commented out the inference function and made the UI update ten thousand times, the value of the Text widget on the APK interface never changed at all. It only refreshed and showed 9999 after all ten thousand updates finished. What a headache.

[yst-one]

Steps/Code to repro the issue?

Run uv run flet debug android directly in the example/yolo_flet_example directory.

[ndonkoHenri]

I can see you use time.sleep(0.1) in your code - that should be the cause of the "It only refreshed and showed 9999 after all ten thousand updates finished." observation, as synchronous sleeping freezes Flet UI. Use await asyncio.sleep(0.1) instead.

Concerning the main concern you have with bytes, I will try to see if I can repro it locally.

[yst-one]

I can see you use time.sleep(0.1) in your code - that should be the cause of the "It only refreshed and showed 9999 after all ten thousand updates finished." observation, as synchronous sleeping freezes Flet UI. Use await asyncio.sleep(0.1) instead.我发现你在代码中使用了 time.sleep(0.1 )。这很可能就是导致“在完成了一万次更新后，屏幕仍然只显示 9999”的原因。因为同步等待会使得 Flet UI 无法正常运行。请改用 await asyncio.sleep(0.1) 吧。

Concerning the main concern you have with bytes, I will try to see if I can repro it locally.关于您所提到的与字节相关的问题，我会试着在本地重现该问题，看看能否确认这个问题确实存在。
Great! asyncio.sleep(0.01) works perfectly.

[FeodorFitsner]

No worries, we are working on a new communication transport for Flet 0.86 which exchanges messages between Python and Dart using Dart ports API - think like very fast in-memory sockets. The design allows program to have any number of parallel ports. Let's say your Flutter widget wants streaming high-quality video frames from a camera. Wrapping those byte arrays into Flet protocol is overkill, as you see. Your widget will be able to create SendPort/ReceivePort pair in Dart and using to talk to Python directly. The overhead is basically copying blocks of memory while sending them through ports. Here's Dart_Bridge project.

Will make a guide once 0.86 is released.