Algorithm 1 Hash Learning algorithm of TEGAH

Input: Training set ${\{X_i,Y_i,Z_i\}}_{i=1}^N$, Binary code length M, Hyper-parameters ∂, Query sets $Quer{y}_i$, Parameters for TEGAH. Output: Binary code $B_{i=x,y,z}^M$, Parameters ${\mathsf{\Theta }}_X$, ${\mathsf{\Theta }}_Y$ and ${\mathsf{\Theta }}_Z$. Initialization: Initialize the parameters ${\mathsf{\Theta }}_X$, ${\mathsf{\Theta }}_Y$ and ${\mathsf{\Theta }}_Z$, maximum iteration number $epoch$, mini-batch size 80. 1:while $iter<epoch$ do 2:    Compute $F_x^k$, $F_y^k$, $F_z^k$ features using Equations (3), (15) and (18) for the training set. 3:    Compute GAT fusion features using Equation (11). 4:    Calculate losses ${\mathcal{L}}_{\mathrm{tri}}$, ${\mathcal{L}}_{\mathrm{class}}$, ${\mathcal{L}}_{\mathrm{quan}}$, ${\mathcal{L}}_{\mathrm{wass}}$ and ${\mathcal{L}}_{\mathrm{kl}}$ using Equations (12), (19), (22), (26) and (27). 5:    Calculate approximate binary hash codes using query sets data. 6:    Input to the trained TEGAH model. 7:    Calculate binary hash codes using the function. 8:end while return the TEGAH model after training.