metrics.py

# Baseline model for "SGCN:Sparse Graph Convolution Network for Pedestrian Trajectory Prediction"
# Source-code directly referred from SGCN at https://github.com/shuaishiliu/SGCN/tree/0ff25cedc04852803787196e83c0bb941d724fc2/metrics.py

import math
import torch
import numpy as np

def ade(predAll,targetAll,count_):
    All = len(predAll)
    sum_all = 0 
    for s in range(All):
        pred = np.swapaxes(predAll[s][:,:count_[s],:],0,1)
        target = np.swapaxes(targetAll[s][:,:count_[s],:],0,1)
        
        N = pred.shape[0]
        T = pred.shape[1]
        sum_ = 0 
        for i in range(N):
            for t in range(T):
                sum_+=math.sqrt((pred[i,t,0] - target[i,t,0])**2+(pred[i,t,1] - target[i,t,1])**2)
        sum_all += sum_/(N*T)
        
    return sum_all/All


def fde(predAll,targetAll,count_):
    All = len(predAll)
    sum_all = 0 
    for s in range(All):
        pred = np.swapaxes(predAll[s][:,:count_[s],:],0,1)
        target = np.swapaxes(targetAll[s][:,:count_[s],:],0,1)
        N = pred.shape[0]
        T = pred.shape[1]
        sum_ = 0 
        for i in range(N):
            for t in range(T-1,T):
                sum_+=math.sqrt((pred[i,t,0] - target[i,t,0])**2+(pred[i,t,1] - target[i,t,1])**2)
        sum_all += sum_/(N)

    return sum_all/All


def seq_to_nodes(seq_,max_nodes = 88):
    seq_ = seq_.squeeze()
    seq_len = seq_.shape[2]
    
    V = np.zeros((seq_len,max_nodes,2))
    for s in range(seq_len):
        step_ = seq_[:,:,s]
        for h in range(len(step_)): 
            V[s,h,:] = step_[h]
            
    return V.squeeze()

def nodes_rel_to_nodes_abs(nodes,init_node):
    nodes_ = np.zeros_like(nodes)
    for s in range(nodes.shape[0]):
        for ped in range(nodes.shape[1]):
            nodes_[s,ped,:] = np.sum(nodes[:s+1,ped,:],axis=0) + init_node[ped,:]

    return nodes_.squeeze()

def closer_to_zero(current,new_v):
    dec =  min([(abs(current),current),(abs(new_v),new_v)])[1]
    if dec != current:
        return True
    else: 
        return False
        
def bivariate_loss(V_pred,V_trgt):
    #mux, muy, sx, sy, corr
    #assert V_pred.shape == V_trgt.shape
    normx = V_trgt[:,:,0]- V_pred[:,:,0]
    normy = V_trgt[:,:,1]- V_pred[:,:,1]

    sx = torch.exp(V_pred[:,:,2]) #sx
    sy = torch.exp(V_pred[:,:,3]) #sy
    corr = torch.tanh(V_pred[:,:,4]) #corr
    
    sxsy = sx * sy

    z = (normx/sx)**2 + (normy/sy)**2 - 2*((corr*normx*normy)/sxsy)
    negRho = 1 - corr**2

    # Numerator
    result = torch.exp(-z/(2*negRho))
    # Normalization factor
    denom = 2 * np.pi * (sxsy * torch.sqrt(negRho))

    # Final PDF calculation
    result = result / denom
    # Numerical stability
    epsilon = 1e-20
    result = -torch.log(torch.clamp(result, min=epsilon))
    result = torch.mean(result)
    
    return result