Depthmap Generator Limitations and Troubleshooting
The laboratory runs a demanding depth model in the user's browser. This keeps image processing on the user's device, but available memory, processing performance, browser support, and ambiguity in the source image remain real limitations.
Mobile reload loops
Most phones allow a browser tab to use less memory than a desktop browser. The decoded source image, depth model, processing tensors, floating-point depth arrays, and multiple canvases can cause the phone's operating system to terminate and reload the page without giving the Lab an error it can display.
Use a desktop or laptop browser, especially for source images larger than 500 x 500 pixels. Repeatedly trying the same image on a phone may cause the same reload loop each time.
The first initialization may be slow
The browser may need to download and prepare the depth model before the first initialization. The Lab attempts to use WebGPU when it is available and falls back to a compatible CPU processing path when necessary. CPU processing can take substantially longer, so leave the page open while the processing status continues to change.
Memory use grows with pixel count
A decoded image and each full-resolution RGBA canvas require roughly four bytes per pixel. Floating-point depth arrays also require roughly four bytes for every value, and several working copies may exist at the same time. The upper read-only preview is limited to 1024 pixels on its longest side, but the editable output and exported PNG can be much larger.
Incorrect foreground and background
The model may interpret shadows as depth, reflections as physical surfaces, transparent objects as part of the background, or overlapping forms in the wrong front-to-back order. Use Depth Palette guidance when source colors intentionally communicate form, then use the brush to correct remaining local errors.
Black and white appear reversed
Check the Depth Convention legend and the Invert Depth setting. The correct convention depends on how the downstream software maps grayscale values to material removal. There is no universal rule that makes black or white deeper in every relief engraving workflow.
The relief and grayscale previews look different
The relief projection applies simulated directional lighting to slopes calculated from neighboring depth values. It can make the shape easier to recognize, but its apparent brightness is not an exact display of grayscale depth. Treat the grayscale preview and exported PNG as the authoritative depth data.
Large exports or browser slowdown
Reduce the linked artwork dimensions or perimeter border padding. Increasing export resolution does not add detail to the depth information inferred by the model. Oversized canvases increase memory use and make every rerender and brush operation more demanding.