Function test

run.py:411–452

(net1, net2)

def test(net1, net2):
    # switch to evaluation mode
    net1.eval()
    net2.eval()
    ptr = 0
    gall_feat_att = np.zeros((ngall, 2048))

    with torch.no_grad():
        for batch_idx, (input, label) in enumerate(gall_loader):
            batch_num = input.size(0)
            input = input.cuda()
            _, feat_att1 = net1(input, input, test_mode[0])
            _, feat_att2 = net2(input, input, test_mode[0])
            feat_att = (feat_att1 + feat_att2) / 2.
            gall_feat_att[ptr:ptr + batch_num, :] = feat_att.detach().cpu().numpy()
            ptr = ptr + batch_num

    # switch to evaluation
    net1.eval()
    net2.eval()
    ptr = 0
    query_feat_att = np.zeros((nquery, 2048))
    with torch.no_grad():
        for batch_idx, (input, label) in enumerate(query_loader):
            batch_num = input.size(0)
            input = input.cuda()
            _, feat_att1 = net1(input, input, test_mode[1])
            _, feat_att2 = net2(input, input, test_mode[1])
            feat_att = (feat_att1 + feat_att2) / 2.
            query_feat_att[ptr:ptr + batch_num, :] = feat_att.detach().cpu().numpy()
            ptr = ptr + batch_num

    # compute the similarity
    distmat_att = np.matmul(query_feat_att, np.transpose(gall_feat_att))

    # evaluation
    if dataset == 'regdb':
        cmc_att, mAP_att, mINP_att = eval_regdb(-distmat_att, query_label, gall_label)
    elif dataset == 'sysu':
        cmc_att, mAP_att, mINP_att = eval_sysu(-distmat_att, query_label, gall_label, query_cam, gall_cam)

    return cmc_att, mAP_att, mINP_att


# training