images

source/3D Location Encoder/NeRF.md

@@ -33,7 +33,7 @@ The `NERFSpatialRelationLocationEncoder` is designed to compute spatial embeddin
 
 ### Features
 <p align="center">
-    <img src="../figs/NeRF.png" alt="NeRF-transformation" title="NeRF-transformation" width="60%" />
+    <img src="../images/NeRF.png" alt="NeRF-transformation" title="NeRF-transformation" width="60%" />
 </p>
 
 ### Configuration Parameters

source/3D Location Encoder/Sphere2Vec-dfs.md

@@ -36,7 +36,7 @@ The `DFTSpatialRelationLocationEncoder` is designed to process spatial relations
 ### Overview
 This position encoder leverages Discrete Fourier Transform (DFT) techniques to encode spatial coordinates into the frequency domain, enabling the model to recognize patterns and relationships that are not immediately apparent in the spatial domain.
     <p align="center">
-      <img src="../figs/dfs.png" alt="dfs-transformation" title="dfs-transformation" width="80%" />
+      <img src="../images/dfs.png" alt="dfs-transformation" title="dfs-transformation" width="80%" />
     </p>
 ### Features
 - **Frequency Domain Conversion**: Transforms spatial data into a frequency-based representation, capturing inherent spatial frequencies and patterns effectively.

source/3D Location Encoder/Sphere2Vec-sphereC+.md

@@ -37,7 +37,7 @@ The `SphereGridSpatialRelationLocationEncoder` is engineered for encoding spatia
 - **Sinusoidal Encoding**: Applies sinusoidal functions to encode spatial differences, enhancing the model's ability to learn from these features.
 - **Configurable Parameters**: Supports customization of encoding parameters such as space dimensionality and computation device.
     <p align="center">
-      <img src="../figs/sphereC+.png" alt="sphereC-plus-transformation" title="sphereC-plus-transformation" width="80%" />
+      <img src="../images/sphereC+.png" alt="sphereC-plus-transformation" title="sphereC-plus-transformation" width="80%" />
     </p>
 ### Configuration Parameters
 - **coord_dim**: Dimensionality of the space being encoded (e.g., 2D, 3D).

source/3D Location Encoder/Sphere2Vec-sphereC.md

@@ -35,7 +35,7 @@ The `SphereSpatialRelationLocationEncoder` is designed for encoding spatial rela
 
 ### Overview
   <p align="center">
-      <img src="../figs/Sphere2Vec-sphereC.png" alt="Sphere2Vec-sphereC-transformation" title="Sphere2Vec-sphereC-transformation" width="60%" />
+      <img src="../images/Sphere2Vec-sphereC.png" alt="Sphere2Vec-sphereC-transformation" title="Sphere2Vec-sphereC-transformation" width="60%" />
   </p>
 #### Spherical Coordinate Transformation
 

source/3D Location Encoder/Sphere2Vec-sphereM+.md

@@ -36,7 +36,7 @@ Processes input coordinates through the location encoder to produce detailed spa
 ### Overview
 This position encoder transforms spatial coordinates using a sophisticated sinusoidal encoding method, featuring multiple scales to capture a wide range of spatial details.
     <p align="center">
-      <img src="../figs/sphereM+.png" alt="sphereM-plus-transformation" title="sphereM-plus-transformation" width="60%" />
+      <img src="../images/sphereM+.png" alt="sphereM-plus-transformation" title="sphereM-plus-transformation" width="60%" />
     </p>
 
 ### Features

source/3D Location Encoder/xyz.md

@@ -47,7 +47,7 @@ Processes a batch of coordinates and converts them into spatial relation embeddi
   - Convert latitude `lat` and longitude `lon` coordinates into radians.
   - Calculate `x, y, z` coordinates using the following equations:
     <p align="center">
-      <img src="../figs/xyz.png" alt="xyz-transformation" title="xyz-transformation" width="80%" />
+      <img src="../images/xyz.png" alt="xyz-transformation" title="xyz-transformation" width="80%" />
     </p>
   - Concatenate `x, y, z` coordinates to form the high-dimensional vector representation.
 

source/Basic Concepts/Single point location encoder.md

@@ -4,16 +4,16 @@ output:
   pdf_document: default
 ---
 
-# 1. Single point location encoder
+# Single point location encoder
 <p align="center">
   <img src="../images/single_location_encoder_structure.png" alt="Location Encoder Structure" title="General Structure of Single Location Encoder" width="30%" />
 </p>
 
 
-## 1.1 EncoderMultiLayerFeedForwardNN()  
+## EncoderMultiLayerFeedForwardNN()  
 `NN(⋅) : ℝ^W -> ℝ^d` is a learnable neural network component which maps the input position embedding `PE(x) ∈ ℝ^W` into the location embedding `Enc(x) ∈ ℝ^d`. A common practice is to define `NN(⋅)` as a multi-layer perceptron, while Mac Aodha et al. (2019) adopted a more complex `NN(⋅)` which includes an initial fully connected layer, followed by a series of residual blocks. The purpose of `NN(⋅)` is to provide a learnable component for the location encoder, which captures the complex interaction between input locations and target labels.
 
-### 1.1.1 Properties
+### Properties
 
 - `input_dim` (int): Dimensionality of the input embeddings.
 - `output_dim` (int): Dimensionality of the output of the network.
@@ -25,7 +25,7 @@ output:
 - `skip_connection` (bool): If set to True, enables skip connections between layers.
 - `context_str` (str, optional): An optional string providing context for this instance, such as indicating its role within a larger model.
 
-### 1.1.3 Methods
+### Methods
 
 #### `__init__(input_dim, output_dim, num_hidden_layers=0, dropout_rate=None, hidden_dim=-1, activation="sigmoid", use_layernormalize=False, skip_connection=False, context_str=None)`
 Constructor for the `EncoderMultiLayerFeedForwardNN` class.
@@ -53,10 +53,10 @@ Defines the forward pass of the network.
 
 
 
-## 1.2 PositionEncoder()
+## PositionEncoder()
 `PE(⋅)` is the most important component which distinguishes different `Enc(x)`. Usually, `PE(⋅)` is a *deterministic* function which transforms location x into a W-dimension vector, so-called position embedding. The purpose of `PE(⋅)` is to do location feature normalization (Chu et al. 2019, Mac Aodha et al. 2019, Rao et al. 2020) and/or feature decomposition (Mai et al. 2020b, Zhong et al. 2020) so that the output `PE(x)` is more learning-friendly for `NN(⋅)`. In Table 1 we further classify different `Enc(x)` into four sub-categories based on their `PE(⋅)`: discretization-based, direct, sinusoidal, and sinusoidal multi-scale location encoder. Each of them will be discussed in detail below.
 
-### 1.2.1 Properties
+### Properties
 - `spa_embed_dim` (int): The dimension of the output spatial relation embedding.
 - `coord_dim` (int): The dimensionality of space (e.g., 2 for 2D, 3 for 3D).
 - `frequency_num` (int): The number of different frequencies/wavelengths for the sinusoidal functions.
@@ -67,7 +67,7 @@ Defines the forward pass of the network.
 - `device` (str): The device to which tensors will be moved ('cuda' or 'cpu').
 
 
-### 1.2.2 Methods
+### Methods
 ### `get_activation_function(activation, context_str)`
 - **Parameters**:
   - `activation`: A string that specifies the type of activation function to retrieve.
@@ -222,6 +222,6 @@ Computes the embedding for a path between nodes.
 <div style="display:none">
 # 2. Aggregation location encoder
 <p align="center">
-  <img src="./figs/aggregation_location_encoder_structure.png" alt="Structure of Aggregation Location Encoder Structure" title="General Structure of Location Encoder" width="40%" />
+  <img src="./images/aggregation_location_encoder_structure.png" alt="Structure of Aggregation Location Encoder Structure" title="General Structure of Location Encoder" width="40%" />
 </p>
 </div>

source/requirements.txt

@@ -1,5 +1,2 @@
 myst-parser
-renku-sphinx-theme
-sphinx_pdj_theme
-groundwork-sphinx-theme
 sphinx-press-theme