Canonical: https://instain.ru/en/ai/marigold/ Language: en EPFL, HUAWEI Bayer Lab and University of Bologna model # Marigold. Depth, surface normals, and albedo from one image 4.40 / 5 [Directory score](https://instain.ru/en/ai/marigold/#editorial-rating) [Official website ](https://github.com/huawei-bayerlab/marigold-v2) A family of computer vision models. Separate Marigold V2 checkpoints estimate depth, surface normals, or albedo from an image. ![Marigold. Marigold V2 repository with a list of predictions and launch examples.](https://instain.ru/assets/images/marigold-screen-8c0d7f1edc9c-1200.webp) Marigold V2 repository with a list of predictions and launch examples. [Open the source ](https://github.com/huawei-bayerlab/marigold-v2) ## Depth map as a guide for installation Marigold estimates which areas in the image are closer to the camera. In the second version, you can choose a different checkpoint to predict surface orientation or material color without directly accounting for lighting. Relative depth helps separate foreground and background. Surface normals show how visible surfaces are tilted. A single photo leaves the true scale and metric distances undetermined. ## Checking the frame with a showcase Take your own picture of the showcase, where the front rack overlaps the shelf, and the background is visible behind the glass. First, evaluate the overall depth map and compare the order of plans with the source. Then bring the border of the rack, a thin support and glare on the glass. If you want to assess the surface behind the glass, check the specialized control point in the documentation. Its conclusion should be compared with an ordinary map, not relying only on the color of the visualization. Keep the original photo and prediction array nearby. The color picture is convenient for viewing, and the array will be useful in preparing the mask. ## Make the run reproducible The official repository describes running on Linux with CUDA and warns of memory requirements that grow with resolution. Start with a small set of photos from your task. Save the parameters and the name of the control point. This will allow you to understand whether the difference in results is due to the angle, frame size or model chosen. Separately check the conditions of use of base and additional weights. ## Suitable tasks - Evaluate the order of objects in the frame - Prepare a guideline for the mask in depth Features ## Strengths - Separate weights for depth, normal and albedo - Numerical array and prediction visualization Things to consider ## Limitations - Relative map does not give distance in meters - Glass, mirrors and borders require manual inspection ## Selection and conditions Free access Not confirmed Trial period Not confirmed No sign-up required Not confirmed Russian interface Not confirmed Local installation Yes API Not confirmed “Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms. Input formats Image Result Depth map, Image ### Terms Authors publish code, checkpoints and demonstration. The documented local installation of the second version uses Linux and a video card with CUDA. From task to first result ## Practical example Task Evaluate the order of objects in the frame What to prepare A picture with a clear foreground and several objects at different distances. ### Task brief or prompt Build a depth map and compare the relative position of objects with the original image. Leave the source unchanged. ### Expected result Depth map for visual verification. ### How to check the result - Check near and far objects. - Separately, look at the transparent surfaces and thin contours. An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted. Check details ## Official sources - [Marigold V2 models and launch instructions ](https://github.com/huawei-bayerlab/marigold-v2) - [Research by Marigold V2 authors ](https://arxiv.org/abs/2609.08084) - [Official project of the first version of Marigold ](https://github.com/prs-eth/Marigold) This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use. About the tool Developer EPFL, HUAWEI Bayer Lab and University of Bologna Category [Images](https://instain.ru/en/category/images/) Access Authors publish code, checkpoints and demonstration. The documented local installation of the second version uses Linux and a video card with CUDA. Payment Code and control points are available in the project repository. Check the code and weights conditions, including those of the base model, and consider the cost of calculations. API No separate API specified [Developer terms ](https://github.com/huawei-bayerlab/marigold-v2) Directory score 4.40 / 5 Features 4.60 / 5 Workflows 4.10 / 5 Ecosystem 4.40 / 5 Documentation 4.50 / 5 Scores are an editorial guide to the directory. They are not user reviews or the results of personal testing. [See the methodology ](https://instain.ru/en/methodology/) Keep exploring ## For similar tasks [Full category ](https://instain.ru/en/category/images/) Images ### [Depth Anything](https://instain.ru/en/ai/depth-anything/) A depth map from a single photo 4.33 / 5 model Free access confirmed