test.py

"""
python test.py -m ./weights/ --cfg_file
"""
from __future__ import print_function
import os
import argparse
import torch
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
import numpy as np
from utils.anchor.prior_box import PriorBox
from utils.ops.nms.nms_wrapper import nms
import cv2
from utils.bbox.box_utils import decode, decode_ldmk
from utils.misc import Timer
from tensorboardX import SummaryWriter
import os

os.environ["CUDA_VISIBLE_DEVICES"] = "3"
writer = SummaryWriter('./test_log/')

target_generator = FCOSTargetGenerator()
parser = argparse.ArgumentParser(description='FaceBoxes')

parser.add_argument('-m', '--trained_model', default='weights/RetinaNet/epoch_180.pth',
                    type=str, help='Trained state_dict file path to open')
parser.add_argument('--cfg_file', default='./configs/retinanet.py', type=str, help='model config file')
parser.add_argument('--cuda', default=True, type=bool, help='Use cuda to train model')
parser.add_argument('--cpu', default=False, type=bool, help='Use cpu nms')
parser.add_argument('--dataset', default='WIDER_challenge', type=str, choices=['AFW', 'PASCAL', 'FDDB', 'WIDER',
                                                                               'WIDER__challenge'], help='dataset')
parser.add_argument('--confidence_threshold', default=0.05, type=float, help='confidence_threshold')
parser.add_argument('--top_k', default=5000, type=int, help='top_k')
parser.add_argument('--nms_threshold', default=0.3, type=float, help='nms_threshold')
parser.add_argument('--keep_top_k', default=750, type=int, help='keep_top_k')
args = parser.parse_args()


def save_roc_file(image_path, bboxes_scores, output_dir='./eval/preds_stu'):
    """
    Save predicted results, including bbox and score into text file.
    Args:
        image_path (string): file name.
        bboxes_scores (np.array|list): the predicted bboxed and scores, layout
            is (xmin, ymin, xmax, ymax, score)
        output_dir (string): output directory.
    """
    image_name = image_path.split('/')[-1]
    image_class = image_path.split('/')[-2]

    odir = output_dir
    if not os.path.exists(os.path.join(odir, image_class)):
        os.makedirs(os.path.join(odir, image_class))

    ofname = os.path.join(odir, img_name + '.txt')
    f = open(ofname, 'w')
    #     f.write('{:s}\n'.format(image_class + '/' + image_name))
    #     f.write('{:d}\n'.format(bboxes_scores.shape[0]))
    for box_score in bboxes_scores:
        xmin, ymin, xmax, ymax, score = box_score
        f.write('{:s} {:.1f} {:.1f} {:.1f} {:.1f} {:.3f}\n'.format(image_name, xmin, ymin, xmax, ymax, score))
    f.close()


def save_widerface_bboxes(image_path, bboxes_scores, output_dir, iteration, landmark_pos=None, draw_landmark=None):
    """
    Save predicted results, including bbox and score into text file.
    Args:
        image_path (string): file name.
        bboxes_scores (np.array|list): the predicted bboxed and scores, layout
            is (xmin, ymin, xmax, ymax, score)
        output_dir (string): output directory.
    """
    image_name = image_path.split('/')[-1]
    image_class = image_path.split('/')[-2]

    odir = os.path.join(output_dir, image_class)
    if not os.path.exists(odir):
        os.makedirs(odir)

    ofname = os.path.join(odir, '%s.txt' % (image_name[:-4]))
    f = open(ofname, 'w')
    f.write('{:s}\n'.format(image_class + '/' + image_name))
    f.write('{:d}\n'.format(bboxes_scores.shape[0]))

    img_origin = np.float32(cv2.imread(image_path, cv2.IMREAD_COLOR))
    #     print(image_path)
    # import pdb
    # pdb.set_trace()
    # lanp.concatenate(landmarks,bboxes_scores[:,-1], axis=1)

    if draw_landmark:
        for idx, landmark in enumerate(landmark_pos):
            landmark = landmark.reshape(5,2)
            score = bboxes_scores[idx, -1]
            if score > 0.4:
                for point in landmark:
                    # import pdb
                    # pdb.set_trace()
                    # print('point',point)
                    cv2.circle(img_origin,(int(point[0]), int(point[1])), radius=2, color=(0, 0, 255), thickness=2)

    for box_score in bboxes_scores:
        xmin, ymin, xmax, ymax, score = box_score
        f.write('{:.1f} {:.1f} {:.1f} {:.1f} {:.3f}\n'.format(xmin, ymin, (
                xmax - xmin + 1), (ymax - ymin + 1), score))
        if score > 0.4:
            if xmax - xmin < 25 and ymax - ymin < 25:
                cv2.rectangle(img_origin, (xmin, ymin), (xmax, ymax), (0, 255, 0), 2)
                # 图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细
                cv2.putText(img_origin, str('%0.2f' % score), (xmin, ymin), cv2.FONT_HERSHEY_COMPLEX, 0.8, (0, 255, 0),
                            1)
            elif xmax - xmin < 35 and ymax - ymin < 35:
                cv2.rectangle(img_origin, (xmin, ymin), (xmax, ymax), (255, 0, 255), 2)
                # 图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细
                cv2.putText(img_origin, str('%0.2f' % score), (xmin, ymin), cv2.FONT_HERSHEY_COMPLEX, 0.8,
                            (255, 0, 255), 1)
            else:
                cv2.rectangle(img_origin, (xmin, ymin), (xmax, ymax), (255, 0, 0), 2)
                # 图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细
                cv2.putText(img_origin, str('%0.2f' % score), (xmin, ymin), cv2.FONT_HERSHEY_COMPLEX, 1, (255, 0, 0), 1)

    # img_write_path = os.path.join('./eval', 'write_img_with_bbox_landmark', image_class)
    img_write_path = os.path.join('./eval', 'write_img_with_bbox', image_class)
    if not os.path.exists(img_write_path):
        os.makedirs(img_write_path)
    img_write_name = os.path.join(img_write_path, image_name)
    cv2.imwrite(img_write_name, img_origin)

    img_origin = cv2.cvtColor(img_origin, cv2.COLOR_BGR2RGB) / 255
    writer.add_image('img_with_pred_box', img_origin, \
                     global_step=iteration, dataformats='HWC')

    f.close()


def check_keys(model, pretrained_state_dict):
    ckpt_keys = set(pretrained_state_dict.keys())
    model_keys = set(model.state_dict().keys())
    used_pretrained_keys = model_keys & ckpt_keys
    unused_pretrained_keys = ckpt_keys - model_keys
    missing_keys = model_keys - ckpt_keys
    print('Missing keys:{}'.format(len(missing_keys)))
    print('Unused checkpoint keys:{}'.format(len(unused_pretrained_keys)))
    print('Used keys:{}'.format(len(used_pretrained_keys)))
    assert len(used_pretrained_keys) > 0, 'load NONE from pretrained checkpoint'
    return True


def remove_prefix(state_dict, prefix):
    ''' Old style model is stored with all names of parameters sharing common prefix 'module.' '''
    print('remove prefix \'{}\''.format(prefix))
    f = lambda x: x.split(prefix, 1)[-1] if x.startswith(prefix) else x
    return {f(key): value for key, value in state_dict.items()}


def load_model(model, pretrained_path):
    print('Loading pretrained model from {}'.format(pretrained_path))
    device = torch.cuda.current_device()
    pretrained_dict = torch.load(pretrained_path, map_location=lambda storage, loc: storage.cuda(device))
    if "state_dict" in pretrained_dict.keys():
        pretrained_dict = remove_prefix(pretrained_dict['state_dict'], 'module.')
    else:
        pretrained_dict = remove_prefix(pretrained_dict, 'module.')
    check_keys(model, pretrained_dict)
    model.load_state_dict(pretrained_dict, strict=False)
    return model


if __name__ == '__main__':
    # net and model
    from mmcv import Config

    cfg = Config.fromfile(args.cfg_file)
    save_folder = os.path.join('./eval/', args.dataset, cfg['test_cfg']['save_folder'])

    import models

    net = models.__dict__[cfg['net_cfg']['net_name']](phase='test', cfg=cfg['net_cfg'])
    net = load_model(net, args.trained_model)

    net.eval()
    print('Finished loading model!')
    print(net)
    if args.cuda:
        net = net.cuda()
        cudnn.benchmark = True
    else:
        net = net.cpu()

    # save file
    if not os.path.exists(save_folder):
        os.makedirs(save_folder)
    if args.dataset != 'WIDER':
        fw = open(os.path.join(save_folder, args.dataset + '_dets.txt'), 'w')

    # testing dataset
    if 'WIDER' in args.dataset:
        testset_folder = os.path.join('/home/gyt/dataset', args.dataset, 'JPEGImages/')
        testset_list = os.path.join('/home/gyt/dataset', args.dataset, 'ImageSets/Main/test.txt')
    elif args.dataset == 'FDDB':
        testset_folder = os.path.join('/home/gyt/dataset', args.dataset, 'JPEGImages/')
        testset_list = os.path.join('/home/gyt/dataset', args.dataset, 'ImageSets/Main/test.txt')


    with open(testset_list, 'r') as fr:
        test_dataset = fr.read().split()
    num_images = len(test_dataset)

    # testing scale
    if args.dataset == "FDDB":
        resize = 2.5
    elif args.dataset == "PASCAL":
        resize = 2.5
    elif args.dataset == "AFW" or "WIDER" in args.dataset:
        resize = 1

    _t = {'forward_pass': Timer(), 'misc': Timer()}

    # name_list = ["0_Parade_Parade_0_519.jpg"]

    # testing begin
    for i, img_name in enumerate(test_dataset):

        iter_idx = i
        image_path = testset_folder + img_name + '.jpg'
        img = np.float32(cv2.imread(image_path, cv2.IMREAD_COLOR))
        origin_scale = torch.Tensor([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])
        landmark_origin_scale = torch.Tensor([img.shape[1], img.shape[0]]).repeat(5)


        if resize != 1:
            img = cv2.resize(img, None, None, fx=resize, fy=resize, interpolation=cv2.INTER_LINEAR)

        im_height, im_width, _ = img.shape
        scale = torch.Tensor([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])
        img -= (104, 117, 123)
        img = img.transpose(2, 0, 1)
        img = torch.from_numpy(img).unsqueeze(0)

        if args.cuda:
            img = img.cuda()
            scale = scale.cuda()
            origin_scale = origin_scale.cuda()
            landmark_origin_scale = landmark_origin_scale.cuda()

        _t['forward_pass'].tic()
        out = net(img)  # forward pass
        _t['forward_pass'].toc()

        _t['misc'].tic()
        #         import pdb
        #         pdb.set_trace()
        if cfg['test_cfg']['is_anchor_base'] is True:
            priorbox = PriorBox(cfg['anchor_cfg'], image_size=(im_height, im_width), phase = 'test')
            priors = priorbox.forward()
            if args.cuda:
                priors = priors.cuda()

            if cfg['train_cfg']['use_landmark']:
                # import pdb
                # pdb.set_trace()
                loc, conf, landmark_delta = out
                # landmarks = decode_landmark(landmark_delta.squeeze(0), priors.data)
                landmarks = decode_ldmk(landmark_delta.squeeze(0), priors.data)

            else:
                loc, conf, _ = out

            boxes = decode(loc.data.squeeze(0), priors.data, cfg['anchor_cfg']['variance'])

        else:
            box_preds, conf, centerness, _ = out

            box_cords = target_generator.generate_box_cords(im_height, im_width)
            if args.cuda:
                box_cords = box_cords.cuda()

            boxes = FCOSBoxConverter(box_preds, box_cords.unsqueeze(0), im_height, im_width).squeeze(0)

        if args.dataset == 'FDDB':
            boxes = boxes * scale / resize
            boxes = boxes.detach().cpu().numpy()
        else:
            if cfg['test_cfg']['is_anchor_base'] is True:
                boxes = boxes * origin_scale
                if cfg['train_cfg']['use_landmark']:
                    landmarks = landmarks * landmark_origin_scale
                    landmarks = landmarks.data.cpu().numpy()
            #                 boxes = boxes * origin_scale / resize
            else:
                boxes = boxes / resize
            boxes = boxes.data.cpu().numpy()

        #         import pdb
        #         pdb.set_trace()
        if cfg['net_cfg']['num_classes'] == 1 and cfg['test_cfg']['is_anchor_base'] == False:
            #             import pdb
            #             pdb.set_trace()
            scores = conf.data.cpu().numpy()[:, 0]
        else:
            scores = conf.data.cpu().numpy()[:, 1]

        # ignore low scores
        inds = np.where(scores > args.confidence_threshold)[0]

        boxes = boxes[inds]
        scores = scores[inds]

        # keep top-K before NMS
        order = scores.argsort()[::-1][:args.top_k]
        boxes = boxes[order]
        scores = scores[order]

        # do NMS
        dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
        dets, nms_idx = nms(dets, args.nms_threshold)
        # keep top-K after NMS
        dets = dets[:args.keep_top_k, :]

        if cfg['train_cfg']['use_landmark']:
            landmarks = landmarks[inds]
            landmarks = landmarks[order]
            landmarks = landmarks[nms_idx]
            landmarks = landmarks[:args.keep_top_k, :]
        else:
            landmarks = None

        _t['misc'].toc()

        # save dets
        if "WIDER" in args.dataset:
            if cfg['train_cfg']['use_landmark']:
                draw_landmark = True
            else:
                draw_landmark = False

            save_widerface_bboxes(image_path, dets, save_folder, iter_idx, landmarks, draw_landmark=draw_landmark)

        #             save_roc_file(image_path, dets)

        if args.dataset == "FDDB":
            fw.write('{:s}\n'.format(img_name))
            fw.write('{:.1f}\n'.format(dets.shape[0]))
            for k in range(dets.shape[0]):
                xmin = dets[k, 0]
                ymin = dets[k, 1]
                xmax = dets[k, 2]
                ymax = dets[k, 3]
                score = dets[k, 4]
                w = xmax - xmin + 1
                h = ymax - ymin + 1
                fw.write('{:.3f} {:.3f} {:.3f} {:.3f} {:.10f}\n'.format(xmin, ymin, w, h, score))

        print('im_detect: {:d}/{:d} forward_pass_time: {:.4f}s misc: {:.4f}s'.format(i + 1, num_images,
                                                                                     _t['forward_pass'].average_time,
                                                                                     _t['misc'].average_time))