update startup screen docs (#265)

Co-authored-by: Benedikt Best <[email protected]>

documentation/basics/startup.md

@@ -18,17 +18,34 @@ It allows to
 When creating a new project (A) the user can choose from several different
 project types:
 
-- **Pixel Classification** can be used to interactively label images and to obtain a pixel prediction on unlabeled images. 
+##### Segmentation Workflows
 
-- **Carving** is useful for segmenting objects in images which cannot be discriminated from each other by appearance but which have a visible boundary.
+- [**Pixel Classification**][pc] can be used to interactively label images and to obtain a pixel prediction on unlabeled images. 
+- [**Autocontext**][autocontext] improves Pixel Classification by running it in two stages, where the second stage uses the probabilities from the first as additional information.
+- [**Neural Network Classification**][nn] allows to run inference with pre-trained convolutional neural networks (CNNs).
+- [**Trainable Domain Adaptation**][tda] combines Pixel Classification with pre-trained convolutional neural networks.
+- In [**Carving**][carving] you can semi-automatically segment objects that are visually similar but have a visible boundary between them.
+- [**Boundary-based Segmentation**][mc] is useful for automatic segmentation of objects in images which cannot be discriminated from each other by appearance but which have a visible boundary.
 
-  - **Carving from pixel predictions** is a workflow where a pixel prediction of an object boundary is used to segment objects.
+##### Object Classification Workflows
 
+- [**Object classification**][oc] can be used to label and classify already segmented objects based on object-level features such as size, orientation, average color etc.
+  - **from binary image** use a _binary object mask_ or _label image_ (e.g. from Cellpose, Stardist) as object input
+  - **from prediction image** use a probability map (generated e.g. with Pixel Classification) as object input
 
-- **Tracking Workflow** allows tracking of a large and unknown number of (possible divisible) objects with similar appearance in 2d+t and 3d+t 
+##### Tracking Workflows
 
-- **Object classification** allows to train and classify already segmented objects in an image using object level features such as size, orientation, average color etc.
-  - **from pixel classification** use a pixel level classifier as first stage to obtain objects
-  - **from binary image** use a binary object mask image as object input
-  - **from prediction image** use a probability map as object input
+- [**Tracking Workflow**][track] allows tracking of a large and unknown number of (possibly divisible) objects with similar appearance in 2d+t and 3d+t
+  - **Animal Tracking** for objects that don't divide
+  - **Tracking** for objects that divide
+  - **Tracking with Learning**, like Tracking, but allows to optimize tracking parameters
 
+
+[autocontext]: {{site.baseurl}}/documentation/autocontext/autocontext.html
+[carving]: {{site.baseurl}}/documentation/carving/carving.html
+[mc]: {{site.baseurl}}/documentation/multicut/multicut.html
+[nn]: {{site.baseurl}}/documentation/nn/nn.html
+[oc]: {{site.baseurl}}/documentation/objects/objects.html
+[pc]: {{site.baseurl}}/documentation/pixelclassification/pixelclassification.html
+[tda]: {{site.baseurl}}/documentation/tda/tda.html
+[track]: {{site.baseurl}}/documentation/tracking/tracking.html