Choose theme
AI Resources
WildDet3D
WildDet3D detects and locates objects in 3D from an RGB image, using text, points, or boxes to specify what to find.
Ai2 publishes the research model, weights, inference guide, and interactive demo. The outputs include 2D and 3D boxes and depth maps, for builders working on spatial perception. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A promptable 3D detection system
An image-to-3D detection model: name an object category or mark an object in the image to request its location and bounding box in 3D.
Why it stands out
Promptable spatial perception
A box prompt can identify one object to lift into 3D, or act as an example for finding similar objects. The selected prompt mode changes which task the model performs.
Availability
Public repo with weights and demos
The official repository includes installation guidance, released model weights, demo materials, application examples, and pointers to local and interactive usage paths.
Why it matters
What makes it useful
For a scene-understanding experiment, you can ask for named objects such as chairs or cars and get their 3D boxes from a single image. The interactive demo provides a way to try those prompts before preparing the local research setup.
What to know
Where it fits
It sits between image input and the application using the detections. Its inference outputs include camera-coordinate 3D boxes in metres; the repository also supplies tracking and vision-language-model integration examples.
Notable points
What stands out
The same drawn box has two different uses. Geometric mode returns one 3D prediction for that box; visual mode treats it as an example and searches for similar objects. Choose the former to locate the marked object, or the latter to find more like it.
Before using
What to review
The local installation uses Python 3.11, PyTorch 2.5.1 with CUDA 12.1, source-built CUDA operations, and recursive submodules. Follow the repository's matched setup.
The repository distinguishes its Stage 3 paper checkpoint from two intermediate training checkpoints; each is about 4.7 GB.
Point and box prompts use the original image's pixel coordinates. The inference wrapper handles resizing and padding.
Ai2 describes the model as intended for research and education under its Responsible Use Guidelines. The model card declares the SAM License; read the linked terms for your intended use.
Check permission and privacy needs for scene inputs containing people or private spaces before uploading or processing them.
Treat the detections as experimental outputs that need testing for your application; the research model is not a finished safety system for robotics, tracking, or AR.
Reader fit
Who may find it relevant
Computer-vision researchers and builders experimenting with image-based 3D object detection.
Teams connecting detections to tracking, robotics, or AR prototypes and able to work with the documented local setup.
The demo offers a lighter first look; local integration requires model and camera-data experience.
Editorial note
Why LifeHubber lists it
For a perception experiment that already has stereo or LiDAR measurements, WildDet3D offers a way to use that measured depth alongside the image instead of relying solely on depth inferred from it. Already have a measured depth map from stereo or LiDAR? The inference guide explains how to use it: preprocess the image with depth in metres at the image's resolution, enable the model's depth-input path, and pass the prepared depth into each inference call. Without those inputs, WildDet3D predicts depth from the image itself.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
More in AI Models
Keep browsing this category
Explore more AI model resources.
Gemma 4
google/gemma-4
A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.
NVIDIA Nemotron 3.5 Super VL Base
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16
NVIDIA vision-language base checkpoint for continued training with text, document images, and video frames, with BF16 weights and a shared context ceiling up to one million tokens.
Lance
bytedance-research/Lance
A ByteDance Research unified multimodal model for image and video understanding, generation, and editing, with model files, demos, inference scripts, Gradio setup, benchmark scripts, and a stated 40GB VRAM inference requirement.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
