melage 0.0.68__py3-none-any.whl → 1.0.0__py3-none-any.whl

melage/plugins/warpseg/warpseg_main/voxelmorph/torch/utils.py

@@ -0,0 +1,470 @@
+import torch
+import torch.nn.functional as F
+import torch.nn as nn
+import ants
+import nibabel as nib
+
+import numpy as np
+from scipy.ndimage import gaussian_filter
+from scipy.ndimage import zoom
+
+def intensity_res(imT1, prob):
+    # Avoid division by zero
+    eps = 1e-5
+
+    # Ensure input tensors are on the same device (e.g., CUDA or CPU)
+    imT1 = imT1.float()  # Ensure imT1 is a float tensor
+    prob = prob.float()  # Ensure prob is a float tensor
+
+    # Broadcast imT1 to match the shape of prob (B, K, D, H, W)
+    if imT1.shape[1] == 1:
+        imT1_broadcast = imT1.expand(-1, prob.shape[1], -1, -1, -1)  # Broadcast imT1 to (B, K, D, H, W)
+    else:
+        imT1_broadcast = imT1  # (B, K, D, H, W), no need to broadcast if shapes already match
+
+    # Mean: mu_k shape (B, K)
+    sum_prob = torch.sum(prob, dim=(2, 3, 4))  # Sum over the spatial dimensions
+    mu_k = torch.sum(prob * imT1_broadcast, dim=(2, 3, 4)) / (sum_prob + eps)
+
+    # Variance: var_k shape (B, K)
+    mu_k_broadcast = mu_k.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)  # (B, K, 1, 1, 1) for broadcasting
+    var_k = torch.sum(prob * (imT1_broadcast - mu_k_broadcast) ** 2, dim=(2, 3, 4)) / (sum_prob + eps)
+
+    # Standard deviation: std_k shape (B, K)
+    std_k = torch.sqrt(var_k + eps)
+
+    # Expand for broadcasting
+    mu = mu_k.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)  # shape (B, K, 1, 1, 1)
+    std = std_k.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)  # shape (B, K, 1, 1, 1)
+
+    # Log-likelihood under Gaussian model
+    log_likelihood = -0.5 * ((imT1_broadcast - mu) ** 2) / (std ** 2 + eps) \
+                     - torch.log(std + eps) \
+                     - 0.5 * np.log(2 * np.pi)
+
+    # Softmax across regions (K) to get confidence weights
+    logits = log_likelihood
+    logits_max = torch.max(logits, dim=1, keepdim=True).values  # Numerical stability (subtract max)
+    exp_logits = torch.exp(logits - logits_max)  # for numerical stability
+    confidence_weights = exp_logits / (torch.sum(exp_logits, dim=1, keepdim=True) + eps)  # shape (B, K, D, H, W)
+
+    return confidence_weights
+
+def create_image_from_prob(imT1, prob_value):
+    topk_probs, topk_labels = torch.topk(prob_value, k=1, dim=1)
+
+    centers = (prob_value * imT1).sum(dim=[2, 3, 4]) / (prob_value.sum(dim=[2, 3, 4]) + 1e-5)  # shape [B, C]
+    # Reshape centers for broadcasting: [B, C, 1, 1, 1]
+    centers_expanded = centers.view(centers.shape[0], centers.shape[1], 1, 1, 1)
+    # Expand to match spatial dims of topk_labels: [B, C, D, H, W]
+    centers_expanded = centers_expanded.expand(-1, -1, *topk_labels.shape[2:])
+    # Make sure topk_labels is shaped for indexing: [B, 1, D, H, W] (for class/channel selection)
+    # It should be long/int type
+    topk_labels = topk_labels.long()  # Add channel dim for gather
+    # Gather values: [B, 1, D, H, W]
+    new_image = torch.gather(centers_expanded, dim=1, index=topk_labels)
+
+    return new_image
+
+def write_to_file(prob_atlas_size,mask_original, mask_new, prob_new,  pathOrigin):
+    imA = read_atlas_file(pathOrigin, rewrite_to_file=True)
+
+    mask_original = mask_original.squeeze().detach().cpu().numpy()
+    mask_original = nib.Nifti1Image(mask_original.astype(np.int16), affine=imA.affine, header=imA.header)
+
+    mask_original.to_filename(mask_new)
+
+    prob_atlas_size = prob_atlas_size.squeeze().detach().cpu().numpy()
+    prob_atlas_size = nib.Nifti1Image(prob_atlas_size, affine=imA.affine, header=imA.header)
+
+    prob_atlas_size.to_filename(prob_new)
+    print(f'OutProb_original: {prob_new}')
+
+
+
+class Histogram_Matching(nn.Module):
+    def __init__(self, differentiable=True, bins=32, sigma=None):
+        super(Histogram_Matching, self).__init__()
+        self.differentiable = differentiable
+        self.bins = bins
+        self.sigma = sigma if sigma is not None else 5.0
+
+    def forward(self, dst, ref):
+        B, C, D, H, W = dst.shape
+        assert dst.device == ref.device
+
+        hist_dst = self.cal_hist(dst)
+        hist_ref = self.cal_hist(ref)
+
+        tables = self.cal_trans_batch(hist_dst, hist_ref)
+
+        dst_flat = dst.view(B*C, -1) * (self.bins - 1)  # scale intensities
+        dst_flat = dst_flat.unsqueeze(2)  # (B*C, N, 1)
+
+        indices = torch.linspace(0, self.bins-1, self.bins, device=dst.device).view(1, 1, -1)  # (1,1,bins)
+
+        weights = torch.softmax(-((dst_flat - indices)**2) / (2*(self.sigma**2)), dim=-1)  # (B*C, N, bins)
+
+        matched = torch.bmm(weights, tables.unsqueeze(-1)).squeeze(-1)  # (B*C, N)
+
+        matched = matched.view(B, C, D, H, W) / (self.bins - 1)  # back to [0,1]
+
+        return matched
+
+    def cal_hist(self, img):
+        B, C, D, H, W = img.shape
+        if self.differentiable:
+            hists = self.soft_histc_batch(img * (self.bins - 1))
+        else:
+            hists = torch.stack([
+                torch.histc(img[b, c] * (self.bins - 1), bins=self.bins, min=0, max=self.bins-1)
+                for b in range(B) for c in range(C)
+            ])
+        hists = F.normalize(hists.float(), p=1, dim=1)
+        cdfs = torch.cumsum(hists, dim=1)
+        return cdfs
+
+    def soft_histc_batch(self, x):
+        B, C, D, H, W = x.shape
+        x = x.view(B*C, -1)
+
+        delta = 1.0
+        centers = torch.linspace(0, self.bins-1, self.bins, device=x.device).float()
+        centers = centers[None, :, None]  # (1, bins, 1)
+
+        x = x.unsqueeze(1)  # (B*C, 1, N)
+
+        soft_bins = torch.sigmoid((x - centers + delta/2) * self.sigma) - torch.sigmoid((x - centers - delta/2) * self.sigma)
+        soft_hist = soft_bins.sum(dim=2)  # (B*C, bins)
+
+        return soft_hist
+
+    def cal_trans_batch(self, cdf_src, cdf_ref):
+        Bc, bins = cdf_src.shape
+        cdf_src = cdf_src.unsqueeze(2)  # (B*C, bins, 1)
+        cdf_ref = cdf_ref.unsqueeze(1)  # (B*C, 1, bins)
+
+        dist = torch.abs(cdf_src - cdf_ref)  # (B*C, bins, bins)
+
+        mapping = dist.argmin(dim=2)  # (B*C, bins)
+
+        return mapping.float()
+
+def erode_mask(mask, radius=1):
+    kernel = torch.ones((1, 1, 2*radius+1, 2*radius+1, 2*radius+1), device=mask.device)
+    padded = F.pad(mask.float(), [radius]*6, mode='replicate')
+    eroded = F.conv3d(padded, kernel) == kernel.numel()
+    return eroded.float()
+
+def adjust_probs_with_uncertainty(probs, uncertainty, min_temp=1.0, max_temp=5.0):
+    # Normalize uncertainty to [0, 1]
+    norm_unc = (uncertainty - uncertainty.min()) / (uncertainty.max() - uncertainty.min() + 1e-5)
+
+    # Convert normalized uncertainty to temperature (higher temp = more smoothing)
+    temp = min_temp + (max_temp - min_temp) * norm_unc  # shape: (B, 1, D, H, W)
+
+    # To avoid instability, apply temp scaling in log-space
+    log_probs = torch.log(probs + 1e-8)
+    log_probs_scaled = log_probs / temp  # broadcast over channels
+
+    probs_adj = F.softmax(log_probs_scaled, dim=1)
+    return probs_adj
+
+def get_back_to_size_torch(img: torch.Tensor, pad_width):
+    (pad_x, pad_y, pad_z) = pad_width
+    pad_x1, pad_x2 = pad_x
+    pad_y1, pad_y2 = pad_y
+    pad_z1, pad_z2 = pad_z
+
+    unpadded_img = img[
+        ...,
+        pad_x1 : -pad_x2 if pad_x2 > 0 else None,
+        pad_y1 : -pad_y2 if pad_y2 > 0 else None,
+        pad_z1 : -pad_z2 if pad_z2 > 0 else None,
+    ]
+    return unpadded_img
+
+
+def calculate_jacobian_determinant_3d(flow):
+    """Calculates the Jacobian determinant of a 3D deformation field."""
+    B, _, D, H, W = flow.shape
+    device = flow.device
+    grid_z, grid_y, grid_x = torch.meshgrid(torch.arange(D, device=device), torch.arange(H, device=device),
+                                            torch.arange(W, device=device), indexing='ij')
+    grid = torch.stack((grid_x, grid_y, grid_z), dim=0).float()
+    grid = grid.unsqueeze(0).repeat(B, 1, 1, 1, 1)
+    transform_map = grid + flow
+
+    diff_x = transform_map[..., 1:] - transform_map[..., :-1]
+    grad_x = torch.cat([diff_x, diff_x[..., -1:]], dim=-1)
+    diff_y = transform_map[..., 1:, :] - transform_map[..., :-1, :]
+    grad_y = torch.cat([diff_y, diff_y[..., -1:, :]], dim=-2)
+    diff_z = transform_map[:, :, 1:, :, :] - transform_map[:, :, :-1, :, :]
+    grad_z = torch.cat([diff_z, diff_z[:, :, -1:, :, :]], dim=-3)
+
+    J_xx, J_yx, J_zx = grad_x[:, 0, ...], grad_x[:, 1, ...], grad_x[:, 2, ...]
+    J_xy, J_yy, J_zy = grad_y[:, 0, ...], grad_y[:, 1, ...], grad_y[:, 2, ...]
+    J_xz, J_yz, J_zz = grad_z[:, 0, ...], grad_z[:, 1, ...], grad_z[:, 2, ...]
+
+    determinant = (J_xx * (J_yy * J_zz - J_zy * J_yz) - J_xy * (J_yx * J_zz - J_zx * J_yz) + J_xz * (
+                J_yx * J_zy - J_zx * J_yy))
+    return determinant.unsqueeze(1)
+
+
+def spatial_transformer_field(image, deformation_field, mode='bilinear'):
+    device = image.device
+    D, H, W = image.shape[2:]
+    d_coords, h_coords, w_coords = torch.meshgrid(torch.linspace(-1, 1, D), torch.linspace(-1, 1, H), torch.linspace(-1, 1, W), indexing='ij')
+    identity_grid = torch.stack([w_coords, h_coords, d_coords], dim=-1).unsqueeze(0).to(device)
+    scaling = torch.tensor([2 / (W - 1), 2 / (H - 1), 2 / (D - 1)], device=device)
+    deformation_field_normalized = deformation_field.permute(0, 2, 3, 4, 1) * scaling
+    sampling_grid = identity_grid + deformation_field_normalized
+    if mode=='bilinear':
+        warped_image = F.grid_sample(image, sampling_grid, mode='bilinear', padding_mode='border', align_corners=True)
+    elif mode=='nearest':
+        warped_image = warped_image = F.grid_sample(image, sampling_grid, mode='nearest', padding_mode='border', align_corners=True)
+    else:
+        return None
+    return warped_image
+
+def augment_based_on_deform_field(source,  atlas_probs):
+    # Define which ROIs you want to smooth
+
+
+
+    IMAGE_SHAPE = source.shape[2:]
+    DEFORMATION_COARSE_GRID = (4, 2, 4)
+    DEFORMATION_MAGNITUDE = 8.0  # Max displacement in pixels
+    deformation_field = generate_smooth_deformation_field(
+        shape=IMAGE_SHAPE,
+        coarse_grid_shape=DEFORMATION_COARSE_GRID,
+        magnitude=DEFORMATION_MAGNITUDE
+    )
+    print("Calculating Jacobian determinant...")
+    jacobian_map = calculate_jacobian_determinant_3d(deformation_field)
+    y_source = spatial_transformer_field(source, deformation_field)
+    prob_atlas = spatial_transformer_field(atlas_probs, deformation_field, mode='nearest')  # *mask
+    # c. Modulate the intensity of the warped image using the Jacobian
+    print("Modulating intensity based on deformation...")
+    INTENSITY_EFFECT_STRENGTH = 0.8  # How much compression/expansion changes brightness
+    INTENSITY_NOISE_LEVEL = 0.01  # How much random noise to add
+    modulated_image = modulate_intensity_with_jacobian(
+        y_source, prob_atlas, jacobian_map,
+        effect_strength=INTENSITY_EFFECT_STRENGTH,
+        noise_level=INTENSITY_NOISE_LEVEL
+    )
+    return modulated_image, prob_atlas
+
+# --- NEW FUNCTION ---
+def modulate_intensity_with_jacobian(base_image, atlas, jacobian_map, effect_strength=0.5, noise_level=0.1):
+    """
+    Modulates image intensity within ROIs based on local Jacobian values.
+
+    Args:
+        base_image (torch.Tensor): The image to modify (1, 1, D, H, W).
+        atlas (torch.Tensor): Atlas with integer labels for ROIs (1, 1, D, H, W).
+        jacobian_map (torch.Tensor): The Jacobian determinant map (1, 1, D, H, W).
+        effect_strength (float): How strongly the Jacobian affects intensity.
+                                 Positive: compression -> brighter.
+                                 Negative: compression -> darker.
+        noise_level (float): The standard deviation of Gaussian noise to add.
+
+    Returns:
+        torch.Tensor: The image with modulated intensities.
+    """
+    #modulated_image = base_image.clone()
+
+    pixels_in_roi = base_image
+    jacobian_in_roi = jacobian_map
+
+    # 1. Calculate the deterministic intensity modulation factor
+    # (J-1) centers the effect: J=1 -> no change. J<1 (compression) -> negative.
+
+    modulation_factor = 1.0 - (jacobian_in_roi - (jacobian_in_roi.mean()+2*jacobian_in_roi.std())) * effect_strength
+
+    # 2. Create a random noise component
+    random_noise = torch.randn_like(pixels_in_roi) * noise_level
+
+    # 3. Apply the changes
+    modified_pixels = pixels_in_roi * modulation_factor + random_noise
+
+    # Place the modified pixels back into the image
+    modulated_image = modified_pixels
+
+    """
+    
+
+    # Process each ROI label found in the atlas
+    for label in torch.unique(atlas)[1:]:  # Skip background (label 0)
+        mask = (atlas == label)
+
+        # Get the pixels and corresponding Jacobian values for this ROI
+        pixels_in_roi = base_image[mask]
+        jacobian_in_roi = jacobian_map[mask]
+
+        # 1. Calculate the deterministic intensity modulation factor
+        # (J-1) centers the effect: J=1 -> no change. J<1 (compression) -> negative.
+        modulation_factor = 1.0 - (jacobian_in_roi - 1.0) * effect_strength
+
+        # 2. Create a random noise component
+        random_noise = torch.randn_like(pixels_in_roi) * noise_level
+
+        # 3. Apply the changes
+        modified_pixels = pixels_in_roi * modulation_factor + random_noise
+
+        # Place the modified pixels back into the image
+        modulated_image[mask] = modified_pixels
+    """
+    return modulated_image
+def generate_smooth_deformation_field(shape, coarse_grid_shape, magnitude):
+
+    """
+    Generates a smooth, random 3D deformation field.
+
+    This works by creating a very low-resolution random vector field and then
+    upsampling and smoothing it into a high-resolution, plausible deformation.
+
+    Args:
+        shape (tuple): The final shape of the field (D, H, W).
+        coarse_grid_shape (tuple): The low-resolution grid size (e.g., (4, 4, 4)).
+        magnitude (float): A scaling factor for the displacement strength.
+
+    Returns:
+        torch.Tensor: The smooth deformation field of shape (1, 3, D, H, W).
+    """
+    # 1. Create the coarse, random vector field
+    coarse_field = np.random.randn(*coarse_grid_shape, 3) * magnitude
+
+    # 2. Upsample the coarse field to the final shape
+    zoom_factors = [s / cs for s, cs in zip(shape, coarse_grid_shape)]
+
+    # Upsample each displacement channel (x, y, z)
+    upsampled_field = np.zeros((*shape, 3))
+    for i in range(3):
+        upsampled_field[..., i] = zoom(coarse_field[..., i], zoom=zoom_factors, order=3)  # Cubic spline interpolation
+
+    # 3. Smooth the upsampled field to make it physically plausible
+    # A large sigma creates very smooth, large-scale deformations.
+    smoothed_field = np.zeros_like(upsampled_field)
+    for i in range(3):
+        smoothed_field[..., i] = gaussian_filter(upsampled_field[..., i], sigma=10)
+
+    # 4. Convert to a PyTorch tensor and format for the spatial transformer
+    # The channels need to be in the second dimension (B, C, D, H, W)
+    # The standard for grid_sample is (x, y, z), which corresponds to (W, H, D)
+    #flow_tensor = torch.from_numpy(smoothed_field).permute(3, 0, 1, 2).unsqueeze(0).float()
+
+    # PyTorch flow fields are ordered (x,y,z), corresponding to W,H,D dimensions
+    # Our numpy array was D,H,W,C so we permute C,D,H,W
+    # Let's re-order the numpy array for clarity before converting to tensor
+    smoothed_field_xyz = smoothed_field[..., [2, 1, 0]]  # Reorder to x,y,z
+    flow_tensor = torch.from_numpy(smoothed_field_xyz).permute(3, 0, 1, 2).unsqueeze(0).float()
+
+    return flow_tensor
+
+
+def restore_zero_torch(cropped_data: torch.Tensor, original_shape, min_max, value=0):
+    """
+    Restore the cropped data back to the original shape with padding (default = 0).
+
+    :param cropped_data: The cropped tensor from remove_zero
+    :param original_shape: The shape of the original full image
+    :param min_max: The min/max indices used for cropping
+    :param value: The padding value to fill
+    :return: Restored full-size tensor
+    """
+    if min_max.abs().sum()==0:
+        return cropped_data
+    B, C, _, _, _ = cropped_data.shape
+    W,H, D = original_shape
+    restored = torch.full((B,C,W,H,D), value, dtype=cropped_data.dtype, device=cropped_data.device)
+
+    restored[:,:,
+    min_max[0][0]:min_max[0][1] + 1,
+    min_max[1][0]:min_max[1][1] + 1,
+    min_max[2][0]:min_max[2][1] + 1
+    ] = cropped_data
+
+    return restored
+
+def change_to_original(im_changed, data, batch_num=0):
+    im1 = get_back_to_size_torch(im_changed, data['pad_width'][batch_num])
+    im1 = restore_zero_torch(im1, data['shape_original'][batch_num], data['min_max'][batch_num])
+    return im1
+
+def differentiable_boundary(mask, kernel_size=3):
+    """Returns soft boundary: high when neighbor differs from center."""
+    # Laplacian-like filter via average pooling
+    pad = kernel_size // 2
+    avg_mask = F.avg_pool3d(mask, kernel_size=kernel_size, stride=1, padding=pad)
+    return torch.abs(avg_mask - mask)
+
+
+def laplacian_3d(image):
+    # 3D Laplacian kernel (center - 6 neighbors)
+    kernel = torch.tensor([[[[0, 0, 0],
+                             [0, -1, 0],
+                             [0, 0, 0]],
+
+                            [[0, -1, 0],
+                             [-1, 6, -1],
+                             [0, -1, 0]],
+
+                            [[0, 0, 0],
+                             [0, -1, 0],
+                             [0, 0, 0]]]], dtype=image.dtype, device=image.device)
+
+    kernel = kernel.unsqueeze(0)  # shape: [1, 1, 3, 3, 3]
+
+    # Apply convolution
+    lap = F.conv3d(image, kernel, padding=1)
+    return lap
+def soft_erode(img, iterations=3):
+    for _ in range(iterations):
+        img = -F.max_pool3d(-img, kernel_size=3, stride=1, padding=1)
+    return img
+
+def soft_skeletonize(img, iterations=3):
+    skel = F.relu(img - soft_erode(img, iterations))
+    return skel
+def gradient_3d(img):
+    dz = img[:, :, 1:, :, :] - img[:, :, :-1, :, :]
+    dy = img[:, :, :, 1:, :] - img[:, :, :, :-1, :]
+    dx = img[:, :, :, :, 1:] - img[:, :, :, :, :-1]
+    return dx, dy, dz
+
+def bending_energy_loss(disp):
+    """
+    disp: tensor of shape (B, 3, D, H, W), displacement field
+    """
+    dx = disp[:, :, 2:, 1:-1, 1:-1] - 2 * disp[:, :, 1:-1, 1:-1, 1:-1] + disp[:, :, :-2, 1:-1, 1:-1]
+    dy = disp[:, :, 1:-1, 2:, 1:-1] - 2 * disp[:, :, 1:-1, 1:-1, 1:-1] + disp[:, :, 1:-1, :-2, 1:-1]
+    dz = disp[:, :, 1:-1, 1:-1, 2:] - 2 * disp[:, :, 1:-1, 1:-1, 1:-1] + disp[:, :, 1:-1, 1:-1, :-2]
+    return (dx.pow(2) + dy.pow(2) + dz.pow(2)).mean()
+
+def gm_highest_at_wm_border(P_WM, P_GM, P_BG, P_GM_previous, eps=1e-6):
+    """
+    Computes a soft, differentiable metric for:
+    where WM touches CSF or BG, is GM the highest?
+    """
+
+
+    # Extract individual tissue probs
+    wm = P_WM
+    gm = P_GM
+    bg = P_BG
+
+    # Define WM border using soft boundary detection
+    kernel = torch.zeros((1, 1, 3, 3, 3), device=P_BG.device)
+    offsets = [(0, 1, 1), (1, 0, 1), (1, 1, 0),
+               (1, 1, 2), (1, 2, 1), (2, 1, 1)]
+    for i, j, k in offsets:
+        kernel[0, 0, i, j, k] = 1.0
+    neighbour_wm = F.conv3d((wm > 0.5).float(), kernel, padding=1)/6.  # (B, 1, H, W, D)
+    wm_boundary = (neighbour_wm-wm).abs()
+    # Compute contact with CSF or BG: high where neighboring probs are nonzero
+    csf_or_bg = ( (bg[0]==0)).float()
+    touching = (wm_boundary * csf_or_bg)>0  # soft mask for "touching" regions
+    gm[touching] = P_GM_previous[touching]
+
+    return gm  # in [0, 1], higher = more GM dominance at WM borders

melage/{utils/DispalyIm.py → rendering/DisplayIm.py}

@@ -22,10 +22,10 @@ import math
 from collections import defaultdict
 from sys import platform
 if platform=='darwin':
-    from melage.utils.Shaders_120 import vsrc, fsrc, fsrcPaint, vsrcPaint
+    from melage.rendering.helpers.Shaders_120 import vsrc, fsrc, fsrcPaint, vsrcPaint
 else:
-    from melage.utils.Shaders_330 import vsrc, fsrc, fsrcPaint, vsrcPaint
-from melage.utils.GMM import GaussianMixture
+    from melage.rendering.helpers.Shaders_330 import vsrc, fsrc, fsrcPaint, vsrcPaint
+
 from melage.utils.utils import LargestCC
 from sklearn.mixture import GaussianMixture
 
@@ -970,97 +970,84 @@ class GLWidget(QOpenGLWidget):
         if self.imSlice is None:
             return
 
-        self.UpdatePaintInfo()
         self.makeObject()
 
-
     def UpdatePaintInfo(self):
         """
-        Update infroamtion of the iamge
-        :return:
-        """
-        self.windowAR = self.width() / (self.height()+0.000001) #windows aspect ratio
-        imWidth = self.imWidth
-        imHeight = self.imHeight
-
-        ratioX = imWidth / self.width()
-        ratioY = imHeight / (self.height()+0.000001)
-        if self.width() > self.height():
-            if imWidth <= imHeight:
-
-                self.zoomed_height = imHeight
-                # self.zoomed_height /= self.imAr
-                self.bottom = 0
-                self.top = self.bottom + self.zoomed_height
-
-                # rescale to the image width
-                self.zoomed_width = ratioY * self.width()
-
-                # image left position
-                self.left = imWidth / 2 - (self.zoomed_height * self.windowAR - self.zoomed_width / 2)
-
-                # image right position
-                self.right = self.left + self.zoomed_width
-            else:
-                if ratioX < 1 and ratioY > 1:
-                    self.zoomed_height = imHeight
-                    self.bottom = 0  #
-                    self.top = self.bottom + self.zoomed_height
-
-                    self.zoomed_width = ratioY * self.width()
-                    self.left = imWidth / 2 - (self.zoomed_height * self.windowAR - self.zoomed_width / 2)
-                    self.right = self.left + self.zoomed_width
-
-
-                else:
-                    self.zoomed_width = imWidth
-                    self.left = 0  # -imWidth * (self.windowAR - 1) / 2
-                    self.right = self.left + self.zoomed_width
-
-                    self.zoomed_height = ratioX * self.height()
-                    self.bottom = imHeight / 2 - (self.zoomed_width / self.windowAR - self.zoomed_height / 2)
-                    self.top = self.bottom + self.zoomed_height
+        Update information of the image with a simplified "fit" scaling.
+
+        This centers the image and scales it to fit within the window
+        while preserving its aspect ratio, creating "letterboxing" or
+        "pillarboxing" as needed.
+        """
+        # Add a small epsilon to prevent division by zero
+        epsilon = 0.000001
+
+        window_w = self.width()
+        window_h = self.height() + epsilon
+        img_w = self.imWidth
+        img_h = self.imHeight
+        #print(f"windw w {window_w} and {window_h}")
+        # Handle case where image might not be loaded yet
+        if img_w == 0 or img_h == 0:
+            self.left = 0
+            self.right = window_w
+            self.bottom = 0
+            self.top = window_h
+            return
 
+        img_h += epsilon  # Add epsilon to image height as well
 
-        elif self.width() <= self.height():
+        # 1. Calculate Aspect Ratios
+        window_ar = window_w / window_h
+        image_ar = img_w / img_h
 
-            if imWidth <= imHeight:
-                if ratioX < 1 and ratioY > 1 or ratioX / ratioY < 1:
-                    self.zoomed_height = imHeight
-                    self.bottom = 0
-                    self.top = self.bottom + self.zoomed_height
+        # 2. Compare aspect ratios to determine the limiting dimension
+        if window_ar > image_ar:
+            # --- CASE 1: Window is "wider" than the image ---
+            # The image is "taller" relative to the window.
+            # We are limited by the image's HEIGHT.
+            # This will create "pillarboxing" (black bars on left/right).
 
-                    self.zoomed_width = ratioY * self.width()
+            # The view's height will be the image's height
+            view_height = img_h
+            # The view's width must be calculated from the view's height
+            # to match the window's aspect ratio.
+            view_width = view_height * window_ar
 
-                    self.left = imWidth / 2 - (self.zoomed_height * self.windowAR - self.zoomed_width / 2)
-                    self.right = self.left + self.zoomed_width
+            # Center the view horizontally around the image's center
+            self.bottom = 0
+            self.top = img_h
+            self.left = (img_w / 2.0) - (view_width / 2.0)
+            self.right = (img_w / 2.0) + (view_width / 2.0)
 
 
-                else:
-                    self.zoomed_width = imWidth
-                    self.left = 0
-                    self.right = self.left + self.zoomed_width
+        else:
+            # --- CASE 2: Image is "wider" than the window ---
+            # The image is "wider" relative to the window.
+            # We are limited by the image's WIDTH.
+            # This will create "letterboxing" (black bars on top/bottom).
 
-                    self.zoomed_height = ratioX * self.height()
-                    self.bottom = imHeight / 2 - (self.zoomed_width / self.windowAR - self.zoomed_height / 2)
-                    self.top = self.bottom + self.zoomed_height
+            # The view's width will be the image's width
+            view_width = img_w
+            # The view's height must be calculated from the view's width
+            # to match the window's aspect ratio.
+            view_height = view_width / window_ar
 
+            # Center the view vertically around the image's center
+            self.left = 0
+            self.right = img_w
+            self.bottom = (img_h / 2.0) - (view_height / 2.0)
+            self.top = (img_h / 2.0) + (view_height / 2.0)
 
-            else:
 
-                self.zoomed_width = imWidth
-                self.left = 0
-                self.right = self.left + self.zoomed_width
 
-                self.zoomed_height = ratioX * self.height()
-                #self.zoomed_height = 2*imWidth/self.imAr
+        # These are just for information, not used by glOrtho
+        self.zoomed_width = view_width
+        self.zoomed_height = view_height
 
-                self.bottom = imHeight / 2 - (self.zoomed_width / self.windowAR - self.zoomed_height / 2)
-                self.top = self.bottom + self.zoomed_height
-        #if self.currentWidnowName != 'axial' and self.imType == 't1':
-         #   self.updateScale( self.width()//2, self.height()//2, 1, 0.5)
         self.halfH = self.height() // 2
-        glEnable(GL_NORMALIZE)  # light normalization
+        glEnable(GL_NORMALIZE)
 
     def drawImage(self):
 
@@ -1394,6 +1381,7 @@ class GLWidget(QOpenGLWidget):
     def resizeGL(self, width, height):
         if self.imSlice is None:
             return
+        self.UpdatePaintInfo()
         side = min(width, height)
         #Set viewport
         glViewport((width- side) // 2, (height-side) // 2, self.imWidth,
@@ -1613,7 +1601,7 @@ class GLWidget(QOpenGLWidget):
             else:
                 self._magic_slice = None
         elif self.enabledCircle:
-            # from melage.utils.utils import
+            # from MELAGE.utils.utils import
             if self._NenabledCircle > 0 and self.colorInd != 9876:
 
                 self._center_circle = []
@@ -1950,7 +1938,7 @@ class GLWidget(QOpenGLWidget):
                 polErase = ConvertPToPolygons(self.erasePoints) # convert to polygons
                 self.erasePolygon(polErase)
         elif self.enabledCircle:
-            #from melage.utils.utils import
+            #from MELAGE.utils.utils import
             if self._NenabledCircle>0:
 
                 self._center_circle=[]
@@ -2077,7 +2065,7 @@ class GLWidget(QOpenGLWidget):
                             seg_c = segmented_image[int(yc_mp)-min_whit[0], int(xc_mp)-min_whit[1]]
                             (segmented_image-seg_c)
 
-                            from melage.utils.utils import Threshold_MultiOtsu
+                            from MELAGE.utils.utils import Threshold_MultiOtsu
                             num_class = 3
 
                             ths = Threshold_MultiOtsu(im1, num_class)

melage/{utils → rendering}/glScientific.py

@@ -3,7 +3,7 @@ __author__= 'Bahram Jafrasteh'
 import numpy as np
 import sys
 sys.path.append('../')
-from melage.graphics import GLViewWidget, GLAxisItem, GLScatterPlotItem, GLGridItem, GLVolumeItem, GLPolygonItem
+from melage.rendering import GLViewWidget, GLAxisItem, GLScatterPlotItem, GLGridItem, GLVolumeItem, GLPolygonItem
 from OpenGL.GL import *
 from collections import defaultdict
 from PyQt5.QtWidgets import QApplication, QWidget, QMenu, QAction, QSlider, QFileDialog