yolov8ToEZB/py/get_map.py

import os, torch, cv2, math, tqdm, time, caffe, shutil, argparse
import numpy as np
from quantization import *
from yolo import non_max_suppression, plot_one_box, yolov8layer, letterbox, xywh2xyxy, box_iou
from config import image_base_path, label_base_path, net_shape
PROJECT_NAME = os.environ['PROJECT_NAME']

def clip_boxes(boxes, shape):
    # Clip boxes (xyxy) to image shape (height, width)
    if isinstance(boxes, torch.Tensor):  # faster individually
        boxes[..., 0].clamp_(0, shape[1])  # x1
        boxes[..., 1].clamp_(0, shape[0])  # y1
        boxes[..., 2].clamp_(0, shape[1])  # x2
        boxes[..., 3].clamp_(0, shape[0])  # y2
    else:  # np.array (faster grouped)
        boxes[..., [0, 2]] = boxes[..., [0, 2]].clip(0, shape[1])  # x1, x2
        boxes[..., [1, 3]] = boxes[..., [1, 3]].clip(0, shape[0])  # y1, y2

def scale_boxes(img1_shape, boxes, img0_shape, ratio_pad=None):
    # Rescale boxes (xyxy) from img1_shape to img0_shape
    if ratio_pad is None:  # calculate from img0_shape
        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
    else:
        gain = ratio_pad[0][0]
        pad = ratio_pad[1]

    boxes[..., [0, 2]] -= pad[0]  # x padding
    boxes[..., [1, 3]] -= pad[1]  # y padding
    boxes[..., :4] /= gain
    clip_boxes(boxes, img0_shape)
    return boxes

def process_batch(detections, labels, iouv):
    """
    Return correct prediction matrix
    Arguments:
        detections (array[N, 6]), x1, y1, x2, y2, conf, class
        labels (array[M, 5]), class, x1, y1, x2, y2
    Returns:
        correct (array[N, 10]), for 10 IoU levels
    """
    correct = np.zeros((detections.shape[0], iouv.shape[0])).astype(bool)
    iou = box_iou(labels[:, 1:], detections[:, :4])
    correct_class = labels[:, 0:1] == detections[:, 5]
    for i in range(len(iouv)):
        x = torch.where((iou >= iouv[i]) & correct_class)  # IoU > threshold and classes match
        if x[0].shape[0]:
            matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy()  # [label, detect, iou]
            if x[0].shape[0] > 1:
                matches = matches[matches[:, 2].argsort()[::-1]]
                matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
                # matches = matches[matches[:, 2].argsort()[::-1]]
                matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
            correct[matches[:, 1].astype(int), i] = True
    return torch.tensor(correct, dtype=torch.bool, device=iouv.device)

def smooth(y, f=0.05):
    # Box filter of fraction f
    nf = round(len(y) * f * 2) // 2 + 1  # number of filter elements (must be odd)
    p = np.ones(nf // 2)  # ones padding
    yp = np.concatenate((p * y[0], y, p * y[-1]), 0)  # y padded
    return np.convolve(yp, np.ones(nf) / nf, mode='valid')  # y-smoothed


def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16, prefix=''):
    """ Compute the average precision, given the recall and precision curves.
    Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
    # Arguments
        tp:  True positives (nparray, nx1 or nx10).
        conf:  Objectness value from 0-1 (nparray).
        pred_cls:  Predicted object classes (nparray).
        target_cls:  True object classes (nparray).
        plot:  Plot precision-recall curve at mAP@0.5
        save_dir:  Plot save directory
    # Returns
        The average precision as computed in py-faster-rcnn.
    """

    # Sort by objectness
    i = np.argsort(-conf)
    tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]

    # Find unique classes
    unique_classes, nt = np.unique(target_cls, return_counts=True)
    nc = unique_classes.shape[0]  # number of classes, number of detections

    # Create Precision-Recall curve and compute AP for each class
    px, py = np.linspace(0, 1, 1000), []  # for plotting
    ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
    for ci, c in enumerate(unique_classes):
        i = pred_cls == c
        n_l = nt[ci]  # number of labels
        n_p = i.sum()  # number of predictions
        if n_p == 0 or n_l == 0:
            continue

        # Accumulate FPs and TPs
        fpc = (1 - tp[i]).cumsum(0)
        tpc = tp[i].cumsum(0)

        # Recall
        recall = tpc / (n_l + eps)  # recall curve
        r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0)  # negative x, xp because xp decreases

        # Precision
        precision = tpc / (tpc + fpc)  # precision curve
        p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1)  # p at pr_score

        # AP from recall-precision curve
        for j in range(tp.shape[1]):
            ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
            if plot and j == 0:
                py.append(np.interp(px, mrec, mpre))  # precision at mAP@0.5

    # Compute F1 (harmonic mean of precision and recall)
    f1 = 2 * p * r / (p + r + eps)

    i = smooth(f1.mean(0), 0.1).argmax()  # max F1 index
    p, r, f1 = p[:, i], r[:, i], f1[:, i]
    tp = (r * nt).round()  # true positives
    fp = (tp / (p + eps) - tp).round()  # false positives
    return tp, fp, p, r, f1, ap, unique_classes.astype(int)


def compute_ap(recall, precision):
    """ Compute the average precision, given the recall and precision curves
    # Arguments
        recall:    The recall curve (list)
        precision: The precision curve (list)
    # Returns
        Average precision, precision curve, recall curve
    """

    # Append sentinel values to beginning and end
    mrec = np.concatenate(([0.0], recall, [1.0]))
    mpre = np.concatenate(([1.0], precision, [0.0]))

    # Compute the precision envelope
    mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))

    # Integrate area under curve
    method = 'interp'  # methods: 'continuous', 'interp'
    if method == 'interp':
        x = np.linspace(0, 1, 101)  # 101-point interp (COCO)
        ap = np.trapz(np.interp(x, mrec, mpre), x)  # integrate
    else:  # 'continuous'
        i = np.where(mrec[1:] != mrec[:-1])[0]  # points where x axis (recall) changes
        ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])  # area under curve

    return ap, mpre, mrec

class CustomDataset(BaseDataset):
    def __init__(self, image_path_list):
        super().__init__()
        self.image_path_list = image_path_list

    def __getitem__(self, item):
        print(item, end=' ')
        # read image
        ori_image = cv2.imdecode(np.fromfile(self.image_path_list[item], np.uint8), cv2.IMREAD_COLOR)

        # letterbox
        process_image, ratio, (dw, dh) = letterbox(ori_image, new_shape=net_shape, auto=False)
        # process_image = cv2.resize(ori_image, (640, 640))
        srcnp = cv2.cvtColor(process_image, cv2.COLOR_BGR2RGB)
        srcnp = srcnp.astype(np.float32) / 256
        srcnp = np.array(srcnp)
        srcnp = np.transpose(srcnp, [2,0,1])
        return srcnp

    def __len__(self):
        return len(self.image_path_list)

def get_result_from_caffe(image_path_list, im_shape=(640, 640) ,conf_thres=0.01, iou_thres=0.45):
    prototxt_file = f'{PROJECT_NAME}/model_caffe/model_0.prototxt'
    caffemodel_file = f'{PROJECT_NAME}/model_caffe/model_0.caffemodel'
    
    dataset = CustomDataset(image_path_list)
    print('begin caffe_forward....')
    since = time.time()
    pred = net.src_forward(prototxt_file, caffemodel_file, dataset, 64)
    print(f'\ncaffe_forward finish. using time:{time.time() - since:.5f}.')
    
    out = yolov8layer([pred['output0'].copy(), pred['output1'].copy(), pred['output2'].copy()])
    out = non_max_suppression([torch.from_numpy(out)], conf_thres=conf_thres, iou_thres=iou_thres)
    for i in tqdm.tqdm(range(len(out)), desc='processing scale_boxes.'):
        ori_image = cv2.imread(image_path_list[i])
        out[i][:, :4] = scale_boxes(im_shape, out[i][:, :4], ori_image.shape[:2])
        
        # vis pred result for debug
        # for j in out[i]:
        #     cv2.rectangle(ori_image, (int(j[0]), int(j[1])), (int(j[2]), int(j[3])), (0, 0, 255), thickness=2)
        # cv2.imwrite('./test.png', ori_image)
    return out

def get_result_from_qkqb(image_path_list, im_shape=(640, 640) ,conf_thres=0.01, iou_thres=0.45):
    qk_file = f'{PROJECT_NAME}/model_quantization/checkpoint_quan.qk'
    qb_file = f'{PROJECT_NAME}/model_quantization/checkpoint_quan.qb'
    
    dataset = CustomDataset(image_path_list)
    print('begin easy_forward....')
    since = time.time()
    pred = net.easy_forward(qk_file, qb_file, dataset, 64)
    print(f'easy_forward finish. using time:{time.time() - since:.5f}.')
    
    out = yolov8layer([pred['/model.22/Concat'].copy(), pred['/model.22/Concat_1'].copy(), pred['/model.22/Concat_2'].copy()])
    out = non_max_suppression([torch.from_numpy(out)], conf_thres=conf_thres,iou_thres=iou_thres)
    for i in tqdm.tqdm(range(len(out)), desc='processing scale_boxes.'):
        ori_image = cv2.imdecode(np.fromfile(image_path_list[i], np.uint8), cv2.IMREAD_COLOR)
        out[i][:, :4] = scale_boxes(im_shape, out[i][:, :4], ori_image.shape[:2])
        
        # vis pred result for debug
        # for j in out[i]:
        #     cv2.rectangle(ori_image, (int(j[0]), int(j[1])), (int(j[2]), int(j[3])), (0, 0, 255), thickness=2)
        # cv2.imwrite('./test.png', ori_image)
    return out

def parse_opt():
    parser = argparse.ArgumentParser()
    parser.add_argument('--type', type=str, choices=['caffe', 'qkqb'], required=True, help='caffe or qkqb')
    parser.add_argument('--iou', type=float, default=0.45, help='iou threshold')
    parser.add_argument('--conf', type=float, default=0.01, help='conf threshold')
    parser.add_argument('--save_path', type=str, default=f'{PROJECT_NAME}/map_img_save', help='image save path')
    parser.add_argument('--device', type=int, default=1, help='device')
    opt = parser.parse_known_args()[0]
    return opt
    
if __name__ == '__main__':
    opt = parse_opt()
    net = Net(opt.device)
    
    iouv = torch.linspace(0.5, 0.95, 10)  # iou vector for mAP@0.5:0.95
    niou = iouv.numel()
    stats = []
    
    name_list = [os.path.splitext(i) for i in os.listdir(image_base_path)]
    image_listdir = [f'{image_base_path}/{i[0]}{i[1]}' for i in name_list]
    label_listdir = [f'{label_base_path}/{i[0]}.txt' for i in name_list]
    
    if os.path.exists(opt.save_path):
        shutil.rmtree(opt.save_path)
    os.makedirs(opt.save_path, exist_ok=True)
    
    if opt.type == 'caffe':
        preds = get_result_from_caffe(image_listdir, net_shape, conf_thres=opt.conf, iou_thres=opt.iou)
    elif opt.type == 'qkqb':
        preds = get_result_from_qkqb(image_listdir, net_shape, conf_thres=opt.conf, iou_thres=opt.iou)
    net.release()
    
    for idx, path in enumerate(tqdm.tqdm(name_list)):
        image_path = f'{image_base_path}/{path[0]}{path[1]}'
        label_path = f'{label_base_path}/{path[0]}.txt'
        
        # read and processing
        ori_image = cv2.imread(image_path)
        height, width, _ = ori_image.shape
        
        # read label
        with open(label_path) as f:
            label = f.readlines()
            if len(label) == 0:
                label = np.empty(shape=(0, 5))
            else:
                label = np.array(list(map(lambda x:np.array(x.strip().split(), dtype=np.float), label)))
                label[:, 1:] *= np.array([width, height, width, height])
                label[:, 1:] = xywh2xyxy(label[:, 1:])
        
        pred = preds[idx]
        # vis label result for debug
        for i in label:
            cv2.rectangle(ori_image, (int(i[1]), int(i[2])), (int(i[3]), int(i[4])), (0, 0, 255), thickness=2) # red label
        for i in pred:
            cv2.rectangle(ori_image, (int(i[0]), int(i[1])), (int(i[2]), int(i[3])), (0, 255, 0), thickness=2) # green pred
        cv2.imwrite(f'{opt.save_path}/{path[0]}{path[1]}', ori_image)
        
        nl, npr = label.shape[0], pred.shape[0]
        correct = torch.zeros(npr, niou, dtype=torch.bool)
        if npr == 0:
            if nl:
                stats.append((correct, *torch.zeros((2, 0)), torch.from_numpy(label[:, 0])))
            continue
        
        if nl:
            correct = process_batch(pred, torch.from_numpy(label), iouv)
        stats.append((correct, pred[:, 4], pred[:, 5], torch.from_numpy(label[:, 0])))
    
    stats = [torch.cat(x, 0).cpu().numpy() for x in zip(*stats)]
    tp, fp, p, r, f1, ap, ap_class = ap_per_class(*stats)
    print(f'precision:{p}')
    print(f'recall:{r}')
    print(f'mAP@0.5:{ap[:, 0]}')