Python torch.min() Examples

def iou(source: Tensor, other: Tensor) -> Tensor:
        source, other = source.unsqueeze(dim=-2).repeat(1, 1, other.shape[-2], 1), \
                        other.unsqueeze(dim=-3).repeat(1, source.shape[-2], 1, 1)

        source_area = (source[..., 2] - source[..., 0]) * (source[..., 3] - source[..., 1])
        other_area = (other[..., 2] - other[..., 0]) * (other[..., 3] - other[..., 1])

        intersection_left = torch.max(source[..., 0], other[..., 0])
        intersection_top = torch.max(source[..., 1], other[..., 1])
        intersection_right = torch.min(source[..., 2], other[..., 2])
        intersection_bottom = torch.min(source[..., 3], other[..., 3])
        intersection_width = torch.clamp(intersection_right - intersection_left, min=0)
        intersection_height = torch.clamp(intersection_bottom - intersection_top, min=0)
        intersection_area = intersection_width * intersection_height

        return intersection_area / (source_area + other_area - intersection_area) 

def plot_evolution_results(hyp):  # from utils.utils import *; plot_evolution_results(hyp)
    # Plot hyperparameter evolution results in evolve.txt
    x = np.loadtxt('evolve.txt', ndmin=2)
    f = fitness(x)
    weights = (f - f.min()) ** 2  # for weighted results
    fig = plt.figure(figsize=(12, 10))
    matplotlib.rc('font', **{'size': 8})
    for i, (k, v) in enumerate(hyp.items()):
        y = x[:, i + 5]
        # mu = (y * weights).sum() / weights.sum()  # best weighted result
        mu = y[f.argmax()]  # best single result
        plt.subplot(4, 5, i + 1)
        plt.plot(mu, f.max(), 'o', markersize=10)
        plt.plot(y, f, '.')
        plt.title('%s = %.3g' % (k, mu), fontdict={'size': 9})  # limit to 40 characters
        print('%15s: %.3g' % (k, mu))
    fig.tight_layout()
    plt.savefig('evolve.png', dpi=200) 

def update_global_state(self, beam):
        "Keeps the coverage vector as sum of attentions"
        if len(beam.prev_ks) == 1:
            beam.global_state["prev_penalty"] = beam.scores.clone().fill_(0.0)
            beam.global_state["coverage"] = beam.attn[-1]
            self.cov_total = beam.attn[-1].sum(1)
        else:
            self.cov_total += torch.min(beam.attn[-1],
                                        beam.global_state['coverage']).sum(1)
            beam.global_state["coverage"] = beam.global_state["coverage"] \
                .index_select(0, beam.prev_ks[-1]).add(beam.attn[-1])

            prev_penalty = self.cov_penalty(beam,
                                            beam.global_state["coverage"],
                                            self.beta)
            beam.global_state["prev_penalty"] = prev_penalty 

def plot_results(start=0, stop=0):  # from utils.utils import *; plot_results()
    # Plot training results files 'results*.txt'
    fig, ax = plt.subplots(2, 5, figsize=(14, 7))
    ax = ax.ravel()
    s = ['GIoU', 'Objectness', 'Classification', 'Precision', 'Recall',
         'val GIoU', 'val Objectness', 'val Classification', 'mAP', 'F1']
    for f in sorted(glob.glob('results*.txt') + glob.glob('../../Downloads/results*.txt')):
        results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
        n = results.shape[1]  # number of rows
        x = range(start, min(stop, n) if stop else n)
        for i in range(10):
            y = results[i, x]
            if i in [0, 1, 2, 5, 6, 7]:
                y[y == 0] = np.nan  # dont show zero loss values
            ax[i].plot(x, y, marker='.', label=f.replace('.txt', ''))
            ax[i].set_title(s[i])
            if i in [5, 6, 7]:  # share train and val loss y axes
                ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])

    fig.tight_layout()
    ax[1].legend()
    fig.savefig('results.png', dpi=200) 

def crop_images_random(path='../images/', scale=0.50):  # from utils.utils import *; crop_images_random()
    # crops images into random squares up to scale fraction
    # WARNING: overwrites images!
    for file in tqdm(sorted(glob.glob('%s/*.*' % path))):
        img = cv2.imread(file)  # BGR
        if img is not None:
            h, w = img.shape[:2]

            # create random mask
            a = 30  # minimum size (pixels)
            mask_h = random.randint(a, int(max(a, h * scale)))  # mask height
            mask_w = mask_h  # mask width

            # box
            xmin = max(0, random.randint(0, w) - mask_w // 2)
            ymin = max(0, random.randint(0, h) - mask_h // 2)
            xmax = min(w, xmin + mask_w)
            ymax = min(h, ymin + mask_h)

            # apply random color mask
            cv2.imwrite(file, img[ymin:ymax, xmin:xmax]) 

def plot_images(imgs, targets, paths=None, fname='images.jpg'):
    # Plots training images overlaid with targets
    imgs = imgs.cpu().numpy()
    targets = targets.cpu().numpy()
    # targets = targets[targets[:, 1] == 21]  # plot only one class

    fig = plt.figure(figsize=(10, 10))
    bs, _, h, w = imgs.shape  # batch size, _, height, width
    bs = min(bs, 16)  # limit plot to 16 images
    ns = np.ceil(bs ** 0.5)  # number of subplots

    for i in range(bs):
        boxes = xywh2xyxy(targets[targets[:, 0] == i, 2:6]).T
        boxes[[0, 2]] *= w
        boxes[[1, 3]] *= h
        plt.subplot(ns, ns, i + 1).imshow(imgs[i].transpose(1, 2, 0))
        plt.plot(boxes[[0, 2, 2, 0, 0]], boxes[[1, 1, 3, 3, 1]], '.-')
        plt.axis('off')
        if paths is not None:
            s = Path(paths[i]).name
            plt.title(s[:min(len(s), 40)], fontdict={'size': 8})  # limit to 40 characters
    fig.tight_layout()
    fig.savefig(fname, dpi=200)
    plt.close() 

def plot_results_overlay(start=0, stop=0):  # from utils.utils import *; plot_results_overlay()
    # Plot training results files 'results*.txt', overlaying train and val losses
    s = ['train', 'train', 'train', 'Precision', 'mAP', 'val', 'val', 'val', 'Recall', 'F1']  # legends
    t = ['GIoU', 'Objectness', 'Classification', 'P-R', 'mAP-F1']  # titles
    for f in sorted(glob.glob('results*.txt') + glob.glob('../../Downloads/results*.txt')):
        results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
        n = results.shape[1]  # number of rows
        x = range(start, min(stop, n) if stop else n)
        fig, ax = plt.subplots(1, 5, figsize=(14, 3.5))
        ax = ax.ravel()
        for i in range(5):
            for j in [i, i + 5]:
                y = results[j, x]
                if i in [0, 1, 2]:
                    y[y == 0] = np.nan  # dont show zero loss values
                ax[i].plot(x, y, marker='.', label=s[j])
            ax[i].set_title(t[i])
            ax[i].legend()
            ax[i].set_ylabel(f) if i == 0 else None  # add filename
        fig.tight_layout()
        fig.savefig(f.replace('.txt', '.png'), dpi=200) 

def shem(roi_probs_neg, negative_count, ohem_poolsize):
    """
    stochastic hard example mining: from a list of indices (referring to non-matched predictions),
    determine a pool of highest scoring (worst false positives) of size negative_count*ohem_poolsize.
    Then, sample n (= negative_count) predictions of this pool as negative examples for loss.
    :param roi_probs_neg: tensor of shape (n_predictions, n_classes).
    :param negative_count: int.
    :param ohem_poolsize: int.
    :return: (negative_count).  indices refer to the positions in roi_probs_neg. If pool smaller than expected due to
    limited negative proposals availabel, this function will return sampled indices of number < negative_count without
    throwing an error.
    """
    # sort according to higehst foreground score.
    probs, order = roi_probs_neg[:, 1:].max(1)[0].sort(descending=True)
    select = torch.tensor((ohem_poolsize * int(negative_count), order.size()[0])).min().int()
    pool_indices = order[:select]
    rand_idx = torch.randperm(pool_indices.size()[0])
    return pool_indices[rand_idx[:negative_count].cuda()] 

def intersect(box_a, box_b):
    """ We resize both tensors to [A,B,2] without new malloc:
    [A,2] -> [A,1,2] -> [A,B,2]
    [B,2] -> [1,B,2] -> [A,B,2]
    Then we compute the area of intersect between box_a and box_b.
    Args:
      box_a: (tensor) bounding boxes, Shape: [A,4].
      box_b: (tensor) bounding boxes, Shape: [B,4].
    Return:
      (tensor) intersection area, Shape: [A,B].
    """
    A = box_a.size(0)
    B = box_b.size(0)
    max_xy = torch.min(box_a[:, 2:].unsqueeze(1).expand(A, B, 2),
                       box_b[:, 2:].unsqueeze(0).expand(A, B, 2))
    min_xy = torch.max(box_a[:, :2].unsqueeze(1).expand(A, B, 2),
                       box_b[:, :2].unsqueeze(0).expand(A, B, 2))
    inter = torch.clamp((max_xy - min_xy), min=0)
    return inter[:, :, 0] * inter[:, :, 1] 

def update_critic(self, obs, action, reward, next_obs, not_done, L, step):
        with torch.no_grad():
            _, policy_action, log_pi, _ = self.actor(next_obs)
            target_Q1, target_Q2 = self.critic_target(next_obs, policy_action)
            target_V = torch.min(target_Q1,
                                 target_Q2) - self.alpha.detach() * log_pi
            target_Q = reward + (not_done * self.discount * target_V)

        # get current Q estimates
        current_Q1, current_Q2 = self.critic(obs, action)
        critic_loss = F.mse_loss(current_Q1,
                                 target_Q) + F.mse_loss(current_Q2, target_Q)
        L.log('train_critic/loss', critic_loss, step)


        # Optimize the critic
        self.critic_optimizer.zero_grad()
        critic_loss.backward()
        self.critic_optimizer.step()

        self.critic.log(L, step) 

def huber_loss(error, delta=1.0):
    """
    Args:
        error: Torch tensor (d1,d2,...,dk)
    Returns:
        loss: Torch tensor (d1,d2,...,dk)

    x = error = pred - gt or dist(pred,gt)
    0.5 * |x|^2                 if |x|<=d
    0.5 * d^2 + d * (|x|-d)     if |x|>d
    Ref: https://github.com/charlesq34/frustum-pointnets/blob/master/models/model_util.py
    """
    abs_error = torch.abs(error)
    #quadratic = torch.min(abs_error, torch.FloatTensor([delta]))
    quadratic = torch.clamp(abs_error, max=delta)
    linear = (abs_error - quadratic)
    loss = 0.5 * quadratic**2 + delta * linear
    return loss 

def distance2bbox(points, distance, max_shape=None):
    """Decode distance prediction to bounding box.

    Args:
        points (Tensor): Shape (n, 2), [x, y].
        distance (Tensor): Distance from the given point to 4
            boundaries (left, top, right, bottom).
        max_shape (tuple): Shape of the image.

    Returns:
        Tensor: Decoded bboxes.
    """
    x1 = points[:, 0] - distance[:, 0]
    y1 = points[:, 1] - distance[:, 1]
    x2 = points[:, 0] + distance[:, 2]
    y2 = points[:, 1] + distance[:, 3]
    if max_shape is not None:
        x1 = x1.clamp(min=0, max=max_shape[1] - 1)
        y1 = y1.clamp(min=0, max=max_shape[0] - 1)
        x2 = x2.clamp(min=0, max=max_shape[1] - 1)
        y2 = y2.clamp(min=0, max=max_shape[0] - 1)
    return torch.stack([x1, y1, x2, y2], -1) 

def intersect(box_a, box_b):
    """ We resize both tensors to [A,B,2] without new malloc:
    [A,2] -> [A,1,2] -> [A,B,2]
    [B,2] -> [1,B,2] -> [A,B,2]
    Then we compute the area of intersect between box_a and box_b.
    Args:
      box_a: (tensor) bounding boxes, Shape: [A,4].
      box_b: (tensor) bounding boxes, Shape: [B,4].
    Return:
      (tensor) intersection area, Shape: [A,B].
    """
    A = box_a.size(0)
    B = box_b.size(0)
    max_xy = torch.min(box_a[:, 2:].unsqueeze(1).expand(A, B, 2),
                       box_b[:, 2:].unsqueeze(0).expand(A, B, 2))
    min_xy = torch.max(box_a[:, :2].unsqueeze(1).expand(A, B, 2),
                       box_b[:, :2].unsqueeze(0).expand(A, B, 2))
    inter = torch.clamp((max_xy - min_xy), min=0)
    return inter[:, :, 0] * inter[:, :, 1] 

def boxes_to_masks(boxes, h, w, padding=0.0):
    n = boxes.shape[0]
    boxes = boxes
    x1 = boxes[:, 0]
    y1 = boxes[:, 1]
    x2 = boxes[:, 2]
    y2 = boxes[:, 3]
    b_w = x2 - x1
    b_h = y2 - y1
    x1 = torch.clamp(x1 - 1 - b_w * padding, min=0)
    x2 = torch.clamp(x2 + 1 + b_w * padding, max=w)
    y1 = torch.clamp(y1 - 1 - b_h * padding, min=0)
    y2 = torch.clamp(y2 + 1 + b_h * padding, max=h)

    rows = torch.arange(w, device=boxes.device, dtype=x1.dtype).view(1, 1, -1).expand(n, h, w)
    cols = torch.arange(h, device=boxes.device, dtype=x1.dtype).view(1, -1, 1).expand(n, h, w)

    masks_left = rows >= x1.view(-1, 1, 1)
    masks_right = rows < x2.view(-1, 1, 1)
    masks_up = cols >= y1.view(-1, 1, 1)
    masks_down = cols < y2.view(-1, 1, 1)

    masks = masks_left * masks_right * masks_up * masks_down

    return masks 

def crop_by_box(masks, box, padding=0.0):
    n, h, w = masks.size()

    b_w = box[2] - box[0]
    b_h = box[3] - box[1]
    x1 = torch.clamp(box[0:1] - b_w * padding - 1, min=0)
    x2 = torch.clamp(box[2:3] + b_w * padding + 1, max=w - 1)
    y1 = torch.clamp(box[1:2] - b_h * padding - 1, min=0)
    y2 = torch.clamp(box[3:4] + b_h * padding + 1, max=h - 1)

    rows = torch.arange(w, device=masks.device, dtype=x1.dtype).view(1, 1, -1).expand(n, h, w)
    cols = torch.arange(h, device=masks.device, dtype=x1.dtype).view(1, -1, 1).expand(n, h, w)

    masks_left = rows >= x1.expand(n, 1, 1)
    masks_right = rows < x2.expand(n, 1, 1)
    masks_up = cols >= y1.expand(n, 1, 1)
    masks_down = cols < y2.expand(n, 1, 1)

    crop_mask = masks_left * masks_right * masks_up * masks_down
    return masks * crop_mask.float(), crop_mask 

def poly2bbox(polys):
    """
    without label
    :param polys: (x1, y1, ..., x4, y4) (n, 8)
    :return: boxes: (xmin, ymin, xmax, ymax) (n, 4)
    """
    n = polys.shape[0]
    xs = np.reshape(polys, (n, 4, 2))[:, :, 0]
    ys = np.reshape(polys, (n, 4, 2))[:, :, 1]

    xmin = np.min(xs, axis=1)
    ymin = np.min(ys, axis=1)
    xmax = np.max(xs, axis=1)
    ymax = np.max(ys, axis=1)

    xmin = xmin[:, np.newaxis]
    ymin = ymin[:, np.newaxis]
    xmax = xmax[:, np.newaxis]
    ymax = ymax[:, np.newaxis]

    return np.concatenate((xmin, ymin, xmax, ymax), 1) 

def intersection_area(yx_min1, yx_max1, yx_min2, yx_max2):
    """
    Calculates the intersection area of two lists of bounding boxes.
    :author 申瑞珉 (Ruimin Shen)
    :param yx_min1: The top left coordinates (y, x) of the first list (size [N1, 2]) of bounding boxes.
    :param yx_max1: The bottom right coordinates (y, x) of the first list (size [N1, 2]) of bounding boxes.
    :param yx_min2: The top left coordinates (y, x) of the second list (size [N2, 2]) of bounding boxes.
    :param yx_max2: The bottom right coordinates (y, x) of the second list (size [N2, 2]) of bounding boxes.
    :return: The matrix (size [N1, N2]) of the intersection area.
    """
    ymin1, xmin1 = torch.split(yx_min1, 1, -1)
    ymax1, xmax1 = torch.split(yx_max1, 1, -1)
    ymin2, xmin2 = torch.split(yx_min2, 1, -1)
    ymax2, xmax2 = torch.split(yx_max2, 1, -1)
    max_ymin = torch.max(ymin1.repeat(1, ymin2.size(0)), torch.transpose(ymin2, 0, 1).repeat(ymin1.size(0), 1)) # PyTorch's bug
    min_ymax = torch.min(ymax1.repeat(1, ymax2.size(0)), torch.transpose(ymax2, 0, 1).repeat(ymax1.size(0), 1)) # PyTorch's bug
    height = torch.clamp(min_ymax - max_ymin, min=0)
    max_xmin = torch.max(xmin1.repeat(1, xmin2.size(0)), torch.transpose(xmin2, 0, 1).repeat(xmin1.size(0), 1)) # PyTorch's bug
    min_xmax = torch.min(xmax1.repeat(1, xmax2.size(0)), torch.transpose(xmax2, 0, 1).repeat(xmax1.size(0), 1)) # PyTorch's bug
    width = torch.clamp(min_xmax - max_xmin, min=0)
    return height * width 

def batch_intersection_area(yx_min1, yx_max1, yx_min2, yx_max2):
    """
    Calculates the intersection area of two lists of bounding boxes for N independent batches.
    :author 申瑞珉 (Ruimin Shen)
    :param yx_min1: The top left coordinates (y, x) of the first lists (size [N, N1, 2]) of bounding boxes.
    :param yx_max1: The bottom right coordinates (y, x) of the first lists (size [N, N1, 2]) of bounding boxes.
    :param yx_min2: The top left coordinates (y, x) of the second lists (size [N, N2, 2]) of bounding boxes.
    :param yx_max2: The bottom right coordinates (y, x) of the second lists (size [N, N2, 2]) of bounding boxes.
    :return: The matrics (size [N, N1, N2]) of the intersection area.
    """
    ymin1, xmin1 = torch.split(yx_min1, 1, -1)
    ymax1, xmax1 = torch.split(yx_max1, 1, -1)
    ymin2, xmin2 = torch.split(yx_min2, 1, -1)
    ymax2, xmax2 = torch.split(yx_max2, 1, -1)
    max_ymin = torch.max(ymin1.repeat(1, 1, ymin2.size(1)), torch.transpose(ymin2, 1, 2).repeat(1, ymin1.size(1), 1)) # PyTorch's bug
    min_ymax = torch.min(ymax1.repeat(1, 1, ymax2.size(1)), torch.transpose(ymax2, 1, 2).repeat(1, ymax1.size(1), 1)) # PyTorch's bug
    height = torch.clamp(min_ymax - max_ymin, min=0)
    max_xmin = torch.max(xmin1.repeat(1, 1, xmin2.size(1)), torch.transpose(xmin2, 1, 2).repeat(1, xmin1.size(1), 1)) # PyTorch's bug
    min_xmax = torch.min(xmax1.repeat(1, 1, xmax2.size(1)), torch.transpose(xmax2, 1, 2).repeat(1, xmax1.size(1), 1)) # PyTorch's bug
    width = torch.clamp(min_xmax - max_xmin, min=0)
    return height * width 

def batch_iou_pair(yx_min1, yx_max1, yx_min2, yx_max2, min=float(np.finfo(np.float32).eps)):
    """
    Pairwisely calculates the IoU of two lists (at the same size M) of bounding boxes for N independent batches.
    :author 申瑞珉 (Ruimin Shen)
    :param yx_min1: The top left coordinates (y, x) of the first lists (size [N, M, 2]) of bounding boxes.
    :param yx_max1: The bottom right coordinates (y, x) of the first lists (size [N, M, 2]) of bounding boxes.
    :param yx_min2: The top left coordinates (y, x) of the second lists (size [N, M, 2]) of bounding boxes.
    :param yx_max2: The bottom right coordinates (y, x) of the second lists (size [N, M, 2]) of bounding boxes.
    :return: The lists (size [N, M]) of the IoU.
    """
    yx_min = torch.max(yx_min1, yx_min2)
    yx_max = torch.min(yx_max1, yx_max2)
    size = torch.clamp(yx_max - yx_min, min=0)
    intersect_area = torch.prod(size, -1)
    area1 = torch.prod(yx_max1 - yx_min1, -1)
    area2 = torch.prod(yx_max2 - yx_min2, -1)
    union_area = torch.clamp(area1 + area2 - intersect_area, min=min)
    return intersect_area / union_area 

def intersect(box_a, box_b):
    """ We resize both tensors to [A,B,2] without new malloc:
    [A,2] -> [A,1,2] -> [A,B,2]
    [B,2] -> [1,B,2] -> [A,B,2]
    Then we compute the area of intersect between box_a and box_b.
    Args:
      box_a: (tensor) bounding boxes, Shape: [A,4].
      box_b: (tensor) bounding boxes, Shape: [B,4].
    Return:
      (tensor) intersection area, Shape: [A,B].
    """
    A = box_a.size(0)
    B = box_b.size(0)
    max_xy = torch.min(box_a[:, 2:].unsqueeze(1).expand(A, B, 2),
                       box_b[:, 2:].unsqueeze(0).expand(A, B, 2))
    min_xy = torch.max(box_a[:, :2].unsqueeze(1).expand(A, B, 2),
                       box_b[:, :2].unsqueeze(0).expand(A, B, 2))
    inter = torch.clamp((max_xy - min_xy), min=0)
    return inter[:, :, 0] * inter[:, :, 1] 

def get_best_begin_point_single(coordinate):
    x1 = coordinate[0][0]
    y1 = coordinate[0][1]
    x2 = coordinate[1][0]
    y2 = coordinate[1][1]
    x3 = coordinate[2][0]
    y3 = coordinate[2][1]
    x4 = coordinate[3][0]
    y4 = coordinate[3][1]
    xmin = min(x1, x2, x3, x4)
    ymin = min(y1, y2, y3, y4)
    xmax = max(x1, x2, x3, x4)
    ymax = max(y1, y2, y3, y4)
    combinate = [[[x1, y1], [x2, y2], [x3, y3], [x4, y4]], [[x2, y2], [x3, y3], [x4, y4], [x1, y1]],
                 [[x3, y3], [x4, y4], [x1, y1], [x2, y2]], [[x4, y4], [x1, y1], [x2, y2], [x3, y3]]]
    dst_coordinate = [[xmin, ymin], [xmax, ymin], [xmax, ymax], [xmin, ymax]]
    force = 100000000.0
    force_flag = 0
    for i in range(4):
        temp_force = cal_line_length(combinate[i][0], dst_coordinate[0]) + cal_line_length(combinate[i][1],
                                                                                           dst_coordinate[
                                                                                               1]) + cal_line_length(
            combinate[i][2], dst_coordinate[2]) + cal_line_length(combinate[i][3], dst_coordinate[3])
        if temp_force < force:
            force = temp_force
            force_flag = i
    if force_flag != 0:
        pass
        # print("choose one direction!")
    return  combinate[force_flag] 

def boxlist_iou(boxlist1, boxlist2):
    """Compute the intersection over union of two set of boxes.
    The box order must be (xmin, ymin, xmax, ymax).

    Arguments:
      box1: (BoxList) bounding boxes, sized [N,4].
      box2: (BoxList) bounding boxes, sized [M,4].

    Returns:
      (tensor) iou, sized [N,M].

    Reference:
      https://github.com/chainer/chainercv/blob/master/chainercv/utils/bbox/bbox_iou.py
    """
    if boxlist1.size != boxlist2.size:
        raise RuntimeError(
            "boxlists should have same image size, got {}, {}".format(boxlist1, boxlist2))

    N = len(boxlist1)
    M = len(boxlist2)

    area1 = boxlist1.area()
    area2 = boxlist2.area()

    box1, box2 = boxlist1.bbox, boxlist2.bbox

    lt = torch.max(box1[:, None, :2], box2[:, :2])  # [N,M,2]
    rb = torch.min(box1[:, None, 2:], box2[:, 2:])  # [N,M,2]

    TO_REMOVE = 1

    wh = (rb - lt + TO_REMOVE).clamp(min=0)  # [N,M,2]
    inter = wh[:, :, 0] * wh[:, :, 1]  # [N,M]

    iou = inter / (area1[:, None] + area2 - inter)
    return iou


# implementation from https://github.com/kuangliu/torchcv/blob/master/torchcv/utils/box.py
# with slight modifications 

def _get_concept_groups_masks(self, concept_groups, k):
        if self._concept_groups_masks[k] is None:
            masks = list()
            for cg in concept_groups:
                if isinstance(cg, six.string_types):
                    cg = [cg]
                mask = None
                for c in cg:
                    new_mask = self.taxnomy[k].similarity(self.features[k], c)
                    mask = torch.min(mask, new_mask) if mask is not None else new_mask
                if k == 2 and _apply_self_mask['relate']:
                    mask = do_apply_self_mask(mask)
                masks.append(mask)
            self._concept_groups_masks[k] = torch.stack(masks, dim=0)
        return self._concept_groups_masks[k] 

def filter_most(self, selected, group, concept_groups):
        mask = self._get_concept_groups_masks(concept_groups, 2)

        # mask[x] = \exists y \in selected, greater(y, x)
        mask = torch.min(mask, selected.unsqueeze(-1).unsqueeze(0)).max(dim=-2)[0]
        # -mask[x] = \forall y \in selected, less_eq(y, x)
        mask = torch.min(selected, -mask)
        if torch.is_tensor(group):
            return (mask * group.unsqueeze(1)).sum(dim=0)
        return mask[group] 

def delta2dbbox_v2(Rrois,
                deltas,
                means=[0, 0, 0, 0, 0],
                stds=[1, 1, 1, 1, 1],
                max_shape=None,
                wh_ratio_clip=16 / 1000):
    means = deltas.new_tensor(means).repeat(1, deltas.size(1) // 5)
    stds = deltas.new_tensor(stds).repeat(1, deltas.size(1) // 5)
    denorm_deltas = deltas * stds + means

    dx = denorm_deltas[:, 0::5]
    dy = denorm_deltas[:, 1::5]
    dw = denorm_deltas[:, 2::5]
    dh = denorm_deltas[:, 3::5]
    dangle = denorm_deltas[:, 4::5]

    max_ratio = np.abs(np.log(wh_ratio_clip))
    dw = dw.clamp(min=-max_ratio, max=max_ratio)
    dh = dh.clamp(min=-max_ratio, max=max_ratio)
    Rroi_x = (Rrois[:, 0]).unsqueeze(1).expand_as(dx)
    Rroi_y = (Rrois[:, 1]).unsqueeze(1).expand_as(dy)
    Rroi_w = (Rrois[:, 2]).unsqueeze(1).expand_as(dw)
    Rroi_h = (Rrois[:, 3]).unsqueeze(1).expand_as(dh)
    Rroi_angle = (Rrois[:, 4]).unsqueeze(1).expand_as(dangle)

    gx = dx * Rroi_w * torch.cos(Rroi_angle) \
         - dy * Rroi_h * torch.sin(Rroi_angle) + Rroi_x
    gy = dx * Rroi_w * torch.sin(Rroi_angle) \
         + dy * Rroi_h * torch.cos(Rroi_angle) + Rroi_y
    gw = Rroi_w * dw.exp()
    gh = Rroi_h * dh.exp()

    gangle = (np.pi / 2.) * dangle + Rroi_angle

    if max_shape is not None:
        # TODO: finish it
        pass

    bboxes = torch.stack([gx, gy, gw, gh, gangle], dim=-1).view_as(deltas)
    return bboxes 

def boxlist_partly_overlap(boxlist1, boxlist2):
    if boxlist1.size != boxlist2.size:
        raise RuntimeError(
            "boxlists should have same image size, got {}, {}".format(boxlist1, boxlist2))

    N = len(boxlist1)
    M = len(boxlist2)

    area1 = boxlist1.area()
    area2 = boxlist2.area()

    box1, box2 = boxlist1.bbox, boxlist2.bbox

    lt = torch.max(box1[:, None, :2], box2[:, :2])  # [N,M,2]
    rb = torch.min(box1[:, None, 2:], box2[:, 2:])  # [N,M,2]

    TO_REMOVE = 1

    wh = (rb - lt + TO_REMOVE).clamp(min=0)  # [N,M,2]
    inter = wh[:, :, 0] * wh[:, :, 1]  # [N,M]

    iou = inter / (area1[:, None] + area2 - inter)
    overlap = iou > 0
    not_complete_overlap = (inter - area1[:, None]) * (inter - area2[None, :]) != 0
    partly_overlap = overlap * not_complete_overlap

    return partly_overlap 

def boxlist_overlap(boxlist1, boxlist2):
    if boxlist1.size != boxlist2.size:
        raise RuntimeError(
            "boxlists should have same image size, got {}, {}".format(boxlist1, boxlist2))

    N = len(boxlist1)
    M = len(boxlist2)

    area1 = boxlist1.area()
    area2 = boxlist2.area()

    box1, box2 = boxlist1.bbox, boxlist2.bbox

    lt = torch.max(box1[:, None, :2], box2[:, :2])  # [N,M,2]
    rb = torch.min(box1[:, None, 2:], box2[:, 2:])  # [N,M,2]

    TO_REMOVE = 1

    wh = (rb - lt + TO_REMOVE).clamp(min=0)  # [N,M,2]
    inter = wh[:, :, 0] * wh[:, :, 1]  # [N,M]

    iou = inter / (area1[:, None] + area2 - inter)
    overlap = iou > 0

    return overlap


# TODO redundant, remove 

def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False):
    # Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
    box2 = box2.t()

    # Get the coordinates of bounding boxes
    if x1y1x2y2:
        # x1, y1, x2, y2 = box1
        b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
        b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
    else:
        # x, y, w, h = box1
        b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
        b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
        b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
        b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2

    # Intersection area
    inter_area = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
                 (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)

    # Union Area
    union_area = ((b1_x2 - b1_x1) * (b1_y2 - b1_y1) + 1e-16) + \
                 (b2_x2 - b2_x1) * (b2_y2 - b2_y1) - inter_area

    iou = inter_area / union_area  # iou
    if GIoU:  # Generalized IoU https://arxiv.org/pdf/1902.09630.pdf
        c_x1, c_x2 = torch.min(b1_x1, b2_x1), torch.max(b1_x2, b2_x2)
        c_y1, c_y2 = torch.min(b1_y1, b2_y1), torch.max(b1_y2, b2_y2)
        c_area = (c_x2 - c_x1) * (c_y2 - c_y1) + 1e-16  # convex area
        return iou - (c_area - union_area) / c_area  # GIoU

    return iou 

def compute_ap(recall, precision):
    """ Compute the average precision, given the recall and precision curves.
    Source: https://github.com/rbgirshick/py-faster-rcnn.
    # Arguments
        recall:    The recall curve (list).
        precision: The precision curve (list).
    # Returns
        The average precision as computed in py-faster-rcnn.
    """

    # Append sentinel values to beginning and end
    mrec = np.concatenate(([0.], recall, [min(recall[-1] + 1E-3, 1.)]))
    mpre = np.concatenate(([0.], precision, [0.]))

    # Compute the precision envelope
    for i in range(mpre.size - 1, 0, -1):
        mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])

    # 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 

def wh_iou(box1, box2):
    # Returns the IoU of wh1 to wh2. wh1 is 2, wh2 is nx2
    box2 = box2.t()

    # w, h = box1
    w1, h1 = box1[0], box1[1]
    w2, h2 = box2[0], box2[1]

    # Intersection area
    inter_area = torch.min(w1, w2) * torch.min(h1, h2)

    # Union Area
    union_area = (w1 * h1 + 1e-16) + w2 * h2 - inter_area

    return inter_area / union_area  # iou