SegRef3D Lite
No project Browser autosave ready
Ask AI Tutorial AlignRef2 ↗ GitHub
0 / 0
LINE EDIT
Export

SegRef3D Lite

Browser-based image segmentation, mask refinement & 3D reconstruction

Try a demo dataset

Electron microscopy: EMPIAR-10478, CC0.
Mouse brain microscopy data: Brain Architecture Project (BAP), CC BY-SA 4.0.
Pancreas-CT (Version 2): TCIA, CC BY 3.0 · TCIA data usage policy (retain attribution and policy link when reusing). Data credits and modifications Apple: Zenodo · RabbitCT: Zenodo · CC BY 4.0
Your images stay on your device during normal SegRef3D Lite use.
BROWSER-LOCAL WORKFLOW

Local Processing

Images loaded into SegRef3D Lite are processed locally in your browser.

SegRef3D Lite does not upload your source images to SegRef3D servers.

Masks and autosave data are stored locally in your browser. Exported files are downloaded directly to your device.

Training ZIP is local and excludes DICOM headers, but image data and user-entered object names may still be identifiable.

Exceptions: SegAnything and SegCT/MRI These AI workflows use Google Colab. You must explicitly upload their generated input ZIP, including working images or the source NIfTI, to your own Google Colab runtime.

ローカル処理について

SegRef3D Lite に読み込んだ画像は、通常ブラウザ内で処理され、元画像が SegRef3D のサーバーへ自動的にアップロードされることはありません。

マスクや自動保存データもブラウザ内に保存され、Exportしたファイルは端末へ直接保存されます。

Training ZIPはbrowser内で生成されDICOM headerを含みませんが、画像自体やユーザーが入力したobject名には識別情報が残る可能性があります。

例外：SegAnything／SegCT/MRI
これらのAI機能は Google Colab を使用します。作成された入力ZIP（作業用画像または元NIfTIを含む）は、ユーザー自身が Google Colab ランタイムへ明示的にアップロードします。
LOCAL PROCESSING

Open local images

Your browser may use the word “upload” when asking for folder access. This only gives SegRef3D permission to read the selected files in this browser.

ブラウザには「アップロード」と表示されますが、選択した画像がGitHubや外部サーバーへ送信されることはありません。

Images stay on this device Editing, autosave, and ZIP generation run locally in your browser.
LOCAL MASK IMPORT

Load label masks

Open grayscale label PNGs (0-20) or legacy SegRef3D colored SVG masks. Files stay on this device.

SVG masks must match the loaded working grid. Review the slice mapping before import.

Project ZIP restores editor settings only when Replace is selected.

Import behavior

Replace each matched frame with the imported mask.

Review SVG slice mapping

SVG contains slice masks, not 3D geometry. The loaded source volume supplies the geometry. Check Obj 1 / Obj 2 and anatomical alignment after import.


        

          
          
        

      

    

    
      

        

          
          

            RESET PROJECT MASKS
            
Clear all masks?

          

        

        

          
All label masks in this project will be cleared.

        

        

          This action cannot be undone
          Mask edit history, drawn lines, and browser autosave masks will also be removed.
        

        

          
          
        

      

    

    
      

        

          

            OBJECT MANAGEMENT
            
Manage Object

          

          
        

        

          

            
            Obj 1
          

          
          
          

            
            
            
          

          

            Remove this object's mask pixels from every frame.
            
          

        

      

    

    
      

        

          

            PROJECT VALIDATION
            
Project Check

          

          
        

        
Run a project check to review this sequence.

        

      

    

    
      

        

          

            AI SEGMENTATION WORKFLOW
            
SegAnything

            Prompt-based segmentation for structures you specify using SAM
          

          
        

        
No AI tracking setup yet.

        
    
          
  1. 
            1
            
    Prepare AI TrackingSet objects, ranges, and box prompts.
    
            
          
    2. 
          
  2. 
            2
            
    Create Input ZIP<source-folder>_seganything_request.zip includes the working image sequence and SegAnything job settings. It stays local until you upload it in Colab.
    
            
          
    3. 
          
  3. 
            3
            
    Run SAM2Open the Colab backend and upload the input ZIP.
    
            
              Open SegAnything
            
          
    4. 
          
  4. 
            4
            
    Load ResultReturn the downloaded label masks to this project.
    
            
          
    5. 
        

        

          

            SegCT/MRI
            Automatic anatomical segmentation with TotalSegmentator
          

          
Load a compatible CT/MRI DICOM series or NIfTI volume to begin.

          
          
Available structures depend on the imaging modality and model. The catalog includes supported open-license CT and MRI structures.

          
          
          

            
            
          

          

            Selected structures
            

          

          
          

            
            Open SegCT/MRI
            
          

          
<source-folder>_segct_mri_request.zip contains the source NIfTI volume and job settings. It stays in your browser until you explicitly upload it to your own Google Colab runtime.

          
        

        

          

            Custom Model
            Predict one target with a TrainRef3D model
          

          
Need a model?

          

            Open TrainRef3D
            Evaluate Model
          

          
To create a new request, select a TrainRef3D Model ZIP. Existing InferRef3D predictions can be imported without re-selecting the model.

          

            
Target
—

            
Input
—

            
Model ID
—

          

          

            
            
            Open InferRef3D
            
          

          
Model validation, request creation, and result import stay in this browser. The request contains source voxels that may be identifiable. Model and Request ZIPs leave the browser only when you explicitly upload both to your own Google Colab runtime.

          
Prediction is an algorithmic segmentation intended for review and correction. It is not an independent clinical diagnosis.

          
          
        

        
      

    

    
      

        
PROCESSING LOCATION CHANGES
SegCT/MRI uses Google Colab

        

          
The exported request ZIP contains the source NIfTI volume and selected structure settings.

          
You must explicitly upload it to your own Google Colab runtime for TotalSegmentator processing. It is not uploaded to a SegRef3D-operated server.

          
Confirm that Google Colab use is permitted by your institution's research data and privacy policies.

        

        

      

    

    
      

        
EXISTING MASK DATA
Choose how to import

        
One or more destination objects already contain mask data.
Replace clears those destination objects first. Merge preserves existing labeled pixels and fills only background pixels.

        

      

    

    
      

        
PROCESSING LOCATION CHANGES
InferRef3D uses Google Colab

        
The Model ZIP contains weights. The Inference Request ZIP contains canonical source voxels and may remain identifiable even without DICOM headers.
You must explicitly upload both ZIPs to your own Google Colab runtime. Nothing is uploaded to a SegRef3D-operated server.
Confirm that Google Colab use is permitted by your institution.

        

      

    

    
      

        
PREDICTION IMPORT
Choose how to import the prediction

        
The target object already contains mask data.
Replace clears only the target object, then applies the prediction. Merge keeps it and adds prediction only to background. Both modes protect every other object from overlap.

        

      

    

    
      

        

          
          

            PROCESSING LOCATION CHANGES
            
SegAnything uses Google Colab

          

        

        

          
SegRef3D Lite normally processes images locally in your browser.

          
SegAnything is different: <source-folder>_seganything_request.zip contains the working image sequence and must be uploaded to your Google Colab runtime for SAM-based processing.

          
Your images are not uploaded to a SegRef3D-operated server.

        

        

          Check your institution's policy
          Make sure uploading research or medical data to Google Colab is permitted under your institution's research data and privacy policies.
        

        

          
SegAnything は Google Colab を使用します

          
SegRef3D Liteでは通常、画像はブラウザ内で処理されます。ただし SegAnything では、作業画像を含む入力ZIPをユーザー自身の Google Colab ランタイムへアップロードする必要があります。

          
画像が SegRef3D の運営するサーバーへアップロードされることはありません。研究データ・医用画像を使用する場合は、所属機関の規定上、Google Colabへのアップロードが許可されていることを確認してください。

        

        

          
          Continue to SegAnything
        

      

    

    
      

        

          

            SEG ANYTHING SETUP
            
AI Tracking Setup

          

          
        


        

          

            

              
                
ObjectPromptsTracking Range

              
              
            

            
No tracking objects.

          


          

            

              
              
              
              
            

            

              
              Current frame 0 / 0
              
            

            

              

                

                  PROMPTS
                  Box keyframes
                

                
              

              

                

                  
FrameTypeBoxActions

                  
                

                
No prompts. Navigate to a useful frame and add a box.

              

            

          

        


        
      

    

    
      

        

          

            IMAGE AND VOLUME TOOLS
            
Image Tools

          

          

            
            

              
              Frame 0 / 0
              
            

            
          

        


        

          
          
          
          
          
        


        

          

            
            
            
            
          

          

            
          

        


        

        

        

        
      

    

    
      

        

          

            VOLUME SURFACE
            
STL Preview