Algorithm 2 Pseudocode for Procedural Rainfall Augmentation

Input: Original image $I_{orig}$ Image directory D (for batch processing) Number of rain layers $N_{layers}$ Number of rain lines per layer $N_{lines}$ Rain line length range $[l_{min},l_{max}]$ Rain line angle mean ${\mu }_{\theta }$ and std ${\sigma }_{\theta }$ Rain line thickness range $[w_{min},w_{max}]$ Rain line brightness range $[b_{min},b_{max}]$ Motion blur kernel size $k_{mb}$ Transparency (per-layer alpha) range $[{\alpha }_{min},{\alpha }_{max}]$ Output path $O_{out}$ Output: Rain-augmented image $I_{rain}$ Step 1: Generate and Overlay Rain Layers 1.Initialize $I_{rain}=I_{orig}$ 2. For $j=1$ to $N_{layers}$: Create empty mask $M_j$ with same size as $I_{orig}$ For $i=1$ to $N_{lines}$: – Randomly sample start point $(x_{start},y_{start})$ – Randomly sample length $l_i\in [l_{min},l_{max}]$ – Randomly sample angle ${\theta }_i\sim \mathcal{N}({\mu }_{\theta },{\sigma }_{\theta })$ – Compute end point $(x_{end},y_{end})$ by $l_i$ and ${\theta }_i$ – Randomly sample thickness $w_i\in [w_{min},w_{max}]$ – Randomly sample brightness $b_i\in [b_{min},b_{max}]$ – Draw line from $(x_{start},y_{start})$ to $(x_{end},y_{end})$ with width $w_i$, brightness $b_i$ on $M_j$ Apply motion blur with kernel size $k_{mb}$ along ${\theta }_j$ to $M_j$ Randomly sample transparency ${\alpha }_j\in [{\alpha }_{min},{\alpha }_{max}]$ Blend $M_j$ onto $I_{rain}$: $I_{rain}=(1-{\alpha }_j)·I_{rain}+{\alpha }_j·M_j$ Step 2: Batch Processing and Output 1.Repeat the above for each $I_{orig}$ in directory D 2.Save $I_{rain}$ to $O_{out}$ 3.Output the rain-augmented images