import torch import torch.nn as nn import torch.nn.functional as F from huggingface_hub import PyTorchModelHubMixin import numpy as np from PIL import Image class REBNCONV(nn.Module): def __init__(self, in_ch=3, out_ch=3, dirate=1, stride=1): super(REBNCONV, self).__init__() self.conv_s1 = nn.Conv2d( in_ch, out_ch, 3, padding=1 * dirate, dilation=1 * dirate, stride=stride ) self.bn_s1 = nn.BatchNorm2d(out_ch) self.relu_s1 = nn.ReLU(inplace=True) def forward(self, x): hx = x xout = self.relu_s1(self.bn_s1(self.conv_s1(hx))) return xout def _upsample_like(src, tar): src = F.interpolate(src, size=tar.shape[2:], mode="bilinear") return src class RSU7(nn.Module): def __init__(self, in_ch=3, mid_ch=12, out_ch=3, img_size=512): super(RSU7, self).__init__() self.in_ch = in_ch self.mid_ch = mid_ch self.out_ch = out_ch self.rebnconvin = REBNCONV(in_ch, out_ch, dirate=1) self.rebnconv1 = REBNCONV(out_ch, mid_ch, dirate=1) self.pool1 = nn.MaxPool2d(2, stride=2, ceil_mode=True) self.rebnconv2 = REBNCONV(mid_ch, mid_ch, dirate=1) self.pool2 = nn.MaxPool2d(2, stride=2, ceil_mode=True) self.rebnconv3 = REBNCONV(mid_ch, mid_ch, dirate=1) self.pool3 = nn.MaxPool2d(2, stride=2, ceil_mode=True) self.rebnconv4 = REBNCONV(mid_ch, mid_ch, dirate=1) self.pool4 = nn.MaxPool2d(2, stride=2, ceil_mode=True) self.rebnconv5 = REBNCONV(mid_ch, mid_ch, dirate=1) self.pool5 = nn.MaxPool2d(2, stride=2, ceil_mode=True) self.rebnconv6 = REBNCONV(mid_ch, mid_ch, dirate=1) self.rebnconv7 = REBNCONV(mid_ch, mid_ch, dirate=2) self.rebnconv6d = REBNCONV(mid_ch * 2, mid_ch, dirate=1) self.rebnconv5d = REBNCONV(mid_ch * 2, mid_ch, dirate=1) self.rebnconv4d = REBNCONV(mid_ch * 2, mid_ch, dirate=1) self.rebnconv3d = REBNCONV(mid_ch * 2, mid_ch, dirate=1) self.rebnconv2d = REBNCONV(mid_ch * 2, mid_ch, dirate=1) self.rebnconv1d = REBNCONV(mid_ch * 2, out_ch, dirate=1) def forward(self, x): hx = x hxin = self.rebnconvin(hx) hx1 = self.rebnconv1(hxin) hx = self.pool1(hx1) hx2 = self.rebnconv2(hx) hx = self.pool2(hx2) hx3 = self.rebnconv3(hx) hx = self.pool3(hx3) hx4 = self.rebnconv4(hx) hx = self.pool4(hx4) hx5 = self.rebnconv5(hx) hx = self.pool5(hx5) hx6 = self.rebnconv6(hx) hx7 = self.rebnconv7(hx6) hx6d = self.rebnconv6d(torch.cat((hx7, hx6), 1)) hx6dup = _upsample_like(hx6d, hx5) hx5d = self.rebnconv5d(torch.cat((hx6dup, hx5), 1)) hx5dup = _upsample_like(hx5d, hx4) hx4d = self.rebnconv4d(torch.cat((hx5dup, hx4), 1)) hx4dup = _upsample_like(hx4d, hx3) hx3d = self.rebnconv3d(torch.cat((hx4dup, hx3), 1)) hx3dup = _upsample_like(hx3d, hx2) hx2d = self.rebnconv2d(torch.cat((hx3dup, hx2), 1)) hx2dup = _upsample_like(hx2d, hx1) hx1d = self.rebnconv1d(torch.cat((hx2dup, hx1), 1)) return hx1d + hxin class BriaRMBG(nn.Module, PyTorchModelHubMixin): def __init__(self, config: dict = {"in_ch": 3, "out_ch": 1}): super(BriaRMBG, self).__init__() self.config = config # Simplified architecture for fallback self.stage1 = RSU7(3, 32, 64) self.stage2 = RSU7(64, 32, 128) self.stage3 = RSU7(128, 64, 256) self.stage4 = RSU7(256, 128, 512) # Decoder self.stage4d = RSU7(1024, 128, 256) self.stage3d = RSU7(512, 64, 128) self.stage2d = RSU7(256, 32, 64) self.stage1d = RSU7(128, 16, 64) self.side1 = nn.Conv2d(64, 1, 3, padding=1) self.side2 = nn.Conv2d(64, 1, 3, padding=1) self.side3 = nn.Conv2d(128, 1, 3, padding=1) self.side4 = nn.Conv2d(256, 1, 3, padding=1) self.side5 = nn.Conv2d(512, 1, 3, padding=1) self.side6 = nn.Conv2d(256, 1, 3, padding=1) self.outconv = nn.Conv2d(6, 1, 1) def forward(self, x): hx = x # Encoder hx1 = self.stage1(hx) hx = F.max_pool2d(hx1, 2, stride=2, ceil_mode=True) hx2 = self.stage2(hx) hx = F.max_pool2d(hx2, 2, stride=2, ceil_mode=True) hx3 = self.stage3(hx) hx = F.max_pool2d(hx3, 2, stride=2, ceil_mode=True) hx4 = self.stage4(hx) hx = F.max_pool2d(hx4, 2, stride=2, ceil_mode=True) # Decoder hx4d = self.stage4d(torch.cat((hx, hx4), 1)) hx4dup = _upsample_like(hx4d, hx3) hx3d = self.stage3d(torch.cat((hx4dup, hx3), 1)) hx3dup = _upsample_like(hx3d, hx2) hx2d = self.stage2d(torch.cat((hx3dup, hx2), 1)) hx2dup = _upsample_like(hx2d, hx1) hx1d = self.stage1d(torch.cat((hx2dup, hx1), 1)) # Side outputs side1 = self.side1(hx1d) side2 = _upsample_like(self.side2(hx2d), side1) side3 = _upsample_like(self.side3(hx3d), side1) side4 = _upsample_like(self.side4(hx4d), side1) side5 = _upsample_like(self.side5(hx), side1) side6 = _upsample_like(self.side6(hx4d), side1) # Fusion out = self.outconv(torch.cat((side1, side2, side3, side4, side5, side6), 1)) return torch.sigmoid(out) def simple_background_removal(image): """ Simple fallback background removal using edge detection and thresholding """ if isinstance(image, np.ndarray): img = image else: img = np.array(image) # Convert to grayscale gray = np.mean(img, axis=2) # Simple edge detection from scipy import ndimage edges = ndimage.sobel(gray) # Create a simple mask based on intensity mask = np.ones_like(gray) # Simple thresholding - assume foreground is in center and has more edges h, w = gray.shape center_mask = np.zeros_like(gray) center_mask[h//4:3*h//4, w//4:3*w//4] = 1 # Combine edge information with center bias mask = (edges > np.percentile(edges, 70)) * center_mask mask = ndimage.binary_fill_holes(mask) mask = ndimage.gaussian_filter(mask.astype(float), sigma=2) return mask