eval.py

import models
import time
import torch
import math
from tqdm import tqdm
import numpy as np
import pandas as pd
from sklearn.metrics import f1_score, accuracy_score, recall_score, precision_score, average_precision_score, roc_auc_score

def eval_model(cfg, model, val_loader, loss_ce, val_batch_size):
    model.eval()
    outpred_list = []
    gt_label_list = []
    video_list = []
    valLoss = 0
    lossTrainNorm = 0
    print("******** Start Testing. ********")

    with torch.no_grad():  # No need to track gradients during validation
        for i, (_, input, target, binary_label, video_id) in enumerate(tqdm(val_loader, desc="Validation", total=len(val_loader))):
            if i == 0:
                ss_time = time.time()
            
            input = input[:,0]
            varInput = torch.autograd.Variable(input.float().cuda())
            varTarget = torch.autograd.Variable(target.contiguous().cuda())
            var_Binary_Target = torch.autograd.Variable(binary_label.contiguous().cuda())

            logit = model(varInput)
            lossvalue = loss_ce(logit, var_Binary_Target)

            valLoss += lossvalue.item()
            lossTrainNorm += 1
            outpred_list.append(logit[:,0].sigmoid().cpu().detach().numpy())
            gt_label_list.append(varTarget.cpu().detach().numpy())
            video_list.append(video_id)
    
    valLoss = valLoss / lossTrainNorm

    outpred = np.concatenate(outpred_list, 0)
    gt_label = np.concatenate(gt_label_list, 0)
    video_list = np.concatenate(video_list, 0)
    pred_labels = [1 if item > 0.5 else 0 for item in outpred]
    true_labels = np.argmax(gt_label, axis=1)

    pred_accuracy = accuracy_score(true_labels, pred_labels)

    return pred_accuracy, video_list, pred_labels, true_labels, outpred