Performance issue: Slow image switching and label loading lag with large images

[CastleChen339]

Search before asking

• I have searched the X-AnyLabeling Docs and issues and found no similar questions.

Question

我要标注的图像尺寸很大，约6k*6k，单个png文件大小约70M
我的标注文件保存在另一个目录，因此每次打开文件夹后，我都需要通过“更改输出目录”来加载标注数据。
但每次都会卡住很久很久，才能看到标注数据，这是为什么？能否设计一个进度条来查看加载标注数据的进度

Additional

App name: X-AnyLabeling
App version: 3.3.3
Device: CPU

Operating System: Windows-10-10.0.26200-SP0
CPU: Intel64 Family 6 Model 197 Stepping 2, GenuineIntel
CPU Count: 16
RAM: 32G DDR5
Disk: 1T SSD
GPU: None
CUDA: None
Python Version: 3.10.16

PyQt5 Version: None
ONNX Version: None
ONNX Runtime Version: None
ONNX Runtime GPU Version: None
OpenCV Contrib Python Headless Version: None

[CVHub520]

Hey there! @CastleChen339,

Thank you for reporting the performance issues with large images (6k×6k, ~70MB PNG files). We understand the frustration with loading delays when switching images.

Loading very large images into memory will cause temporary CPU pressure, which is a common limitation when working with high-resolution images. We recommend backing up your current images and compressing the dataset using appropriate compression algorithms (e.g., adjusting PNG compression levels or converting to more efficient formats) to reduce file sizes before loading them into X-AnyLabeling. This should improve loading performance and reduce memory usage while keeping the same image resolution.

If you still experience label loading delays after compressing your images, we recommend running X-AnyLabeling from source code, as it may provide better performance optimizations than pre-built binaries.

Also, having too many labels in a single image can cause performance issues, as the canvas needs to render each target object in real time and manage state overhead, which is currently unavoidable. We apologize for any inconvenience.

We are continuously working on performance optimizations, and your feedback is valuable. If you have further questions or suggestions, please let us know.