from typing import overload, Tuple, Optional import os import cv2 import torch from torch import nn import torch.nn.functional as F import torchvision.transforms as T import numpy as np from glob import glob from PIL import Image from einops import rearrange from model.cldm import ControlLDM from model.gaussian_diffusion import Diffusion from model.bsrnet import RRDBNet from model.swinir import SwinIR from model.scunet import SCUNet from utils.sampler import SpacedSampler from utils.cond_fn import Guidance from utils.video_visualizer import VideoVisualizer from utils.common import wavelet_decomposition, wavelet_reconstruction, count_vram_usage import vidtome from GMFlow.gmflow.gmflow import GMFlow from utils.flow_utils import get_warped_and_mask def save_video(input_folder, out_path, output_name, fps=25): video_visualizer = VideoVisualizer(path=os.path.join(out_path, output_name), frame_size=None, fps=fps) input_folder = os.path.join(out_path, input_folder) imgs = sorted([filename for filename in os.listdir(input_folder) if filename.endswith(('.png', '.jpg'))], key=lambda x: int(x.split('.')[0])) for img in imgs: img_pth = os.path.join(input_folder, img) image = cv2.imread(img_pth) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) video_visualizer.add(image) video_visualizer.save() def batch_bicubic_resize(img: np.ndarray, scale: float) -> np.ndarray: if scale != 1: for i in range(img.shape[0]): img[i] = bicubic_resize(img[i], scale) # pil = Image.fromarray(img) # res = pil.resize(tuple(int(x * scale) for x in pil.size), Image.BICUBIC) return img def bicubic_resize(img: np.ndarray, scale: float) -> np.ndarray: if scale != 1: pil = Image.fromarray(img) res = pil.resize(tuple(int(x * scale) for x in pil.size), Image.BICUBIC) return np.array(res) def resize_short_edge_to(imgs: torch.Tensor, size: int) -> torch.Tensor: _, _, h, w = imgs.size() if h == w: new_h, new_w = size, size elif h < w: new_h, new_w = size, int(w * (size / h)) else: new_h, new_w = int(h * (size / w)), size return F.interpolate(imgs, size=(new_h, new_w), mode="bicubic", antialias=True) def pad_to_multiples_of(imgs: torch.Tensor, multiple: int) -> torch.Tensor: _, _, h, w = imgs.size() if h % multiple == 0 and w % multiple == 0: return imgs.clone() # get_pad = lambda x: (x // multiple + 1) * multiple - x get_pad = lambda x: (x // multiple + int(x % multiple != 0)) * multiple - x ph, pw = get_pad(h), get_pad(w) return F.pad(imgs, pad=(0, pw, 0, ph), mode="constant", value=0) class Pipeline: def __init__(self, stage1_model: nn.Module, cldm: ControlLDM, diffusion: Diffusion, cond_fn: Optional[Guidance], device: str) -> None: self.stage1_model = stage1_model self.cldm = cldm self.diffusion = diffusion self.cond_fn = cond_fn self.device = device self.final_size: Tuple[int] = None def set_final_size(self, lq: torch.Tensor) -> None: h, w = lq.shape[2:] self.final_size = (h, w) @overload def run_stage1(self, lq: torch.Tensor) -> torch.Tensor: ... @count_vram_usage def run_stage2( self, clean: torch.Tensor, steps: int, strength: float, tiled: bool, tile_size: int, tile_stride: int, pos_prompt: str, neg_prompt: str, cfg_scale: float, better_start: float, index: int = 0, input: str = None ) -> torch.Tensor: ### preprocess bs, _, ori_h, ori_w = clean.shape # pad: ensure that height & width are multiples of 64 pad_clean = pad_to_multiples_of(clean, multiple=64) h, w = pad_clean.shape[2:] if self.cldm.controller is not None: self.cldm.controller.cldm = self.cldm self.cldm.controller.non_pad_ratio = (ori_h / h, ori_w / w) self.cldm.vae.decoder.controller = self.cldm.controller # prepare conditon if not tiled: cond = self.cldm.prepare_condition(pad_clean, [pos_prompt] * bs) uncond = self.cldm.prepare_condition(pad_clean, [neg_prompt] * bs) else: cond = self.cldm.prepare_condition_tiled(pad_clean, [pos_prompt] * bs, tile_size, tile_stride) uncond = self.cldm.prepare_condition_tiled(pad_clean, [neg_prompt] * bs, tile_size, tile_stride) if self.cond_fn: self.cond_fn.load_target(pad_clean * 2 - 1) old_control_scales = self.cldm.control_scales self.cldm.control_scales = [strength] * 13 if better_start: # using noised low frequency part of condition as a better start point of # reverse sampling, which can prevent our model from generating noise in # image background. _, low_freq = wavelet_decomposition(pad_clean) # low_freq = pad_clean if not tiled: x_0 = self.cldm.vae_encode(low_freq, batch_size=5) else: x_0 = self.cldm.vae_encode_tiled(low_freq, tile_size, tile_stride) x_T = self.diffusion.q_sample( x_0, torch.full((bs, ), self.diffusion.num_timesteps - 1, dtype=torch.long, device=self.device), torch.randn(x_0.shape, dtype=torch.float32, device=self.device) ) # print(f"diffusion sqrt_alphas_cumprod: {self.diffusion.sqrt_alphas_cumprod[-1]}") else: if self.cldm.latent_control: print(f"[INFO] random initialize {bs} same latents") x_T = 1 * torch.randn((1, 4, h // 8, w // 8), dtype=torch.float32, device=self.device) x_T = x_T.repeat(bs, 1, 1, 1) else: print(f"[INFO] random initialize {bs} latents") x_T = torch.randn((bs, 4, h // 8, w // 8), dtype=torch.float32, device=self.device) ''' loaded latents ''' # t = 981 # latent_fname = f'noisy_latents_{t}.pt' # # model_key = config.model_key.split('/')[-1] # model_key = "stable-diffusion-2-1-base" # inversion_path = os.path.join("latents", os.path.basename(input), "latents") # # outputs/bear_4_BD/latents/stable-diffusion-v1-5/noisy_latents_981.pt # lp = os.path.join(inversion_path, model_key, latent_fname) # latents = torch.load(lp) # # init_noise = latents.to(dtype).to(args.device) # x_T = latents[index][None].to(torch.float32).to(self.device) # print(f"[INFO] loaded latents[{index}]") ''' loaded latent ended ''' ### run sampler sampler = SpacedSampler(self.diffusion.betas) z = sampler.sample( model=self.cldm, device=self.device, steps=steps, batch_size=bs, x_size=(4, h // 8, w // 8), cond=cond, uncond=uncond, cfg_scale=cfg_scale, x_T=x_T, progress=True, progress_leave=True, cond_fn=self.cond_fn, tiled=tiled, tile_size=tile_size, tile_stride=tile_stride, non_pad_ratio=(ori_h / h, ori_w / w) ) if not tiled: if ori_w > 1500: x = self.cldm.vae_decode(z, batch_size=2) else: x = self.cldm.vae_decode(z, batch_size=5) else: x = self.cldm.vae_decode_tiled(z, tile_size // 8, tile_stride // 8) ### postprocess self.cldm.control_scales = old_control_scales sample = x[:, :, :ori_h, :ori_w] return sample @torch.no_grad() def run( self, lq: np.ndarray, steps: int, strength: float, tiled: bool, tile_size: int, tile_stride: int, pos_prompt: str, neg_prompt: str, cfg_scale: float, better_start: bool, index: int = 0, input: str = None, final_size: Tuple[int] = None, flow_model: GMFlow = None, hq: np.ndarray = None ) -> np.ndarray: # image to tensor lq = torch.tensor((lq / 255.).clip(0, 1), dtype=torch.float32, device=self.device) lq = rearrange(lq, "n h w c -> n c h w").contiguous() # set pipeline output size if final_size is None: self.set_final_size(lq) else: self.final_size = final_size clean = self.run_stage1(lq) print(f"[INFO] {clean.shape}") # import ipdb; ipdb.set_trace() # clean = F.interpolate(lq, size=clean.shape[-2:], mode='bicubic', align_corners=False) ''' hq flow & occlusion mask ''' # hq = torch.tensor((hq / 255.).clip(0, 1), dtype=torch.float32, device=self.device) # hq = rearrange(hq, "n h w c -> n c h w").contiguous() # hq = resize_short_edge_to(hq, size=512) # pre_keyframe_lq = None # if self.cldm.controller is not None and \ # self.cldm.controller.step_store["pre_keyframe_lq"] is not None: # pre_keyframe_lq = self.cldm.controller.step_store["pre_keyframe_lq"] # pre_keyframe_lq = torch.tensor((pre_keyframe_lq / 255.).clip(0, 1), dtype=torch.float32, device=self.device) # pre_keyframe_lq = rearrange(pre_keyframe_lq, "n h w c -> n c h w").contiguous() # pre_keyframe_lq = resize_short_edge_to(pre_keyframe_lq, size=512) # pre_keyframe_clean = pre_keyframe_lq[0] # # pre_keyframe_clean = self.run_stage1(pre_keyframe_lq)[0] # flows, masks, confids = [], [], [] # mid = lq.shape[0] // 2 # for k in range(lq.shape[0]): # if k == mid: # if pre_keyframe_lq is not None: # tar_img = (torch.clamp(hq[mid], 0 ,1) * 255).float().to(self.device) # src_img = (torch.clamp(pre_keyframe_clean, 0 ,1) * 255).float().to(self.device) # else: # flows.append(None) # masks.append(None) # confids.append(None) # continue # else: # tar_img = (torch.clamp(hq[k], 0 ,1) * 255).float().to(self.device) # src_img = (torch.clamp(hq[mid], 0 ,1) * 255).float().to(self.device) # # tar_img = stage1_x[0].float().to(args.device) # _, bwd_occ, bwd_flow, bwd_confid = get_warped_and_mask( # flow_model, src_img, tar_img, image3=None, pixel_consistency=False, return_confidence=True) # blend_mask = T.GaussianBlur(kernel_size=(9, 9), sigma=(18, 18))( # F.max_pool2d(bwd_occ, kernel_size=9, stride=1, padding=4)) # blend_mask = torch.clamp(blend_mask + bwd_occ, 0, 1) # blend_mask = 1 - F.max_pool2d(blend_mask, kernel_size=8) # bwd_confid = F.max_pool2d(bwd_confid, kernel_size=8) # bwd_flow = F.interpolate(bwd_flow / 8.0, scale_factor=1. / 8, mode='bilinear') # # _, _, h, w = bwd_flow.shape # # bwd_flow = pad_to_multiples_of(bwd_flow, 8) # # padding_ratio = w / bwd_flow.shape[3] # blend_mask = pad_to_multiples_of(blend_mask[None], 8)[0] # # bwd_confid = pad_to_multiples_of(bwd_confid[None], 8)[0] # flows.append(bwd_flow) # masks.append(blend_mask) # confids.append(bwd_confid) # if self.cldm.controller is not None: # self.cldm.controller.set_warp(flows, masks, flow_confids=confids) ''' flow & occlusion mask ''' pre_keyframe_lq = None if self.cldm.controller is not None and \ self.cldm.controller.step_store["pre_keyframe_lq"] is not None: pre_keyframe_lq = self.cldm.controller.step_store["pre_keyframe_lq"] pre_keyframe_lq = torch.tensor((pre_keyframe_lq / 255.).clip(0, 1), dtype=torch.float32, device=self.device) pre_keyframe_lq = rearrange(pre_keyframe_lq, "n h w c -> n c h w").contiguous() pre_keyframe_clean = self.run_stage1(pre_keyframe_lq)[0] flows, masks, confids = [], [], [] flows2, confids2 = [], [] mid = lq.shape[0] // 2 for k in range(lq.shape[0]): if k == mid: if pre_keyframe_lq is not None: tar_img = (torch.clamp(clean[mid], 0 ,1) * 255).float().to(self.device) src_img = (torch.clamp(pre_keyframe_clean, 0 ,1) * 255).float().to(self.device) else: flows.append(None) masks.append(None) confids.append(None) continue else: tar_img = (torch.clamp(clean[k], 0 ,1) * 255).float().to(self.device) src_img = (torch.clamp(clean[mid], 0 ,1) * 255).float().to(self.device) # tar_img = stage1_x[0].float().to(args.device) _, bwd_occ, bwd_flow, bwd_confid = get_warped_and_mask( flow_model, src_img, tar_img, image3=None, pixel_consistency=False, return_confidence=True) blend_mask = T.GaussianBlur(kernel_size=(9, 9), sigma=(18, 18))( F.max_pool2d(bwd_occ, kernel_size=9, stride=1, padding=4)) blend_mask = torch.clamp(blend_mask + bwd_occ, 0, 1) blend_mask = 1 - F.max_pool2d(blend_mask, kernel_size=8) blend_mask = 1 - F.max_pool2d(bwd_occ, kernel_size=8) bwd_confid2 = F.max_pool2d(bwd_confid, kernel_size=16) bwd_flow2 = F.interpolate(bwd_flow / 16.0, scale_factor=1. / 16, mode='bilinear') bwd_confid = F.max_pool2d(bwd_confid, kernel_size=8) bwd_flow = F.interpolate(bwd_flow / 8.0, scale_factor=1. / 8, mode='bilinear') # _, _, h, w = bwd_flow.shape # bwd_flow = pad_to_multiples_of(bwd_flow, 8) # padding_ratio = w / bwd_flow.shape[3] blend_mask = pad_to_multiples_of(blend_mask[None], 8)[0] # bwd_confid = pad_to_multiples_of(bwd_confid[None], 8)[0] flows.append(bwd_flow) masks.append(blend_mask) confids.append(bwd_confid) flows2.append(bwd_flow2) confids2.append(bwd_confid2) if self.cldm.controller is not None: self.cldm.controller.set_warp(flows, masks, flow_confids=confids) # import ipdb; ipdb.set_trace() _, H, W = confids[0].shape self.cldm.controller.set_flow_correspondence(lq.shape[0], H, W, lq.shape[0] // 2, confids, flows) _, H, W = confids2[0].shape self.cldm.controller.set_flow_correspondence(lq.shape[0], H, W, lq.shape[0] // 2, confids2, flows2) for j, flow in enumerate(self.cldm.controller.step_store["flows"]): if flow is not None: self.cldm.controller.step_store["flows"][j] = pad_to_multiples_of(self.cldm.controller.step_store["flows"][j], 8) # self.cldm.controller.set_warp2(flows2, confids2) ''' flow & occlusion mask ended ''' sample = self.run_stage2( clean, steps, strength, tiled, tile_size, tile_stride, pos_prompt, neg_prompt, cfg_scale, better_start, index=index, input=input ) if self.cldm.controller is not None: print(f"[INFO] clearing controller correspondence scores ... ") self.cldm.controller.step_store["corres_scores"] = None # colorfix (borrowed from StableSR, thanks for their work) sample = (sample + 1) / 2 sample = wavelet_reconstruction(sample, clean) # resize to desired output size sample = F.interpolate(sample, size=self.final_size, mode="bicubic", antialias=True) clean = F.interpolate(clean, size=self.final_size, mode="bilinear", antialias=True) # tensor to image sample = rearrange(sample * 255., "n c h w -> n h w c") sample = sample.contiguous().clamp(0, 255).to(torch.uint8).cpu().numpy() clean = rearrange(clean * 255., "n c h w -> n h w c") clean = clean.contiguous().clamp(0, 255).to(torch.uint8).cpu().numpy() return sample, clean class BSRNetPipeline(Pipeline): def __init__(self, bsrnet: RRDBNet, cldm: ControlLDM, diffusion: Diffusion, cond_fn: Optional[Guidance], device: str, upscale: float) -> None: super().__init__(bsrnet, cldm, diffusion, cond_fn, device) self.upscale = upscale def set_final_size(self, lq: torch.Tensor) -> None: h, w = lq.shape[2:] self.final_size = (int(h * self.upscale), int(w * self.upscale)) @count_vram_usage def run_stage1(self, lq: torch.Tensor) -> torch.Tensor: # NOTE: upscale is always set to 4 in our experiments if lq.shape[-2] > 1000: clean = [] for i in range(lq.shape[0]): torch.cuda.empty_cache() clean.append(self.stage1_model(lq[i:i+1])) clean = torch.cat(clean, dim=0) else: clean = self.stage1_model(lq) # if self.final_size[0] < 512 and self.final_size[1] < 512: if min(self.final_size) < 512: clean = resize_short_edge_to(clean, size=512) else: clean = F.interpolate(clean, size=self.final_size, mode="bicubic", antialias=True) return clean class SwinIRPipeline(Pipeline): def __init__(self, swinir: SwinIR, cldm: ControlLDM, diffusion: Diffusion, cond_fn: Optional[Guidance], device: str) -> None: super().__init__(swinir, cldm, diffusion, cond_fn, device) @count_vram_usage def run_stage1(self, lq: torch.Tensor) -> torch.Tensor: # NOTE: lq size is always equal to 512 in our experiments # resize: ensure the input lq size is as least 512, since SwinIR is trained on 512 resolution if min(lq.shape[2:]) < 512: lq = resize_short_edge_to(lq, size=512) ori_h, ori_w = lq.shape[2:] # pad: ensure that height & width are multiples of 64 pad_lq = pad_to_multiples_of(lq, multiple=64) # run clean = self.stage1_model(pad_lq) # remove padding clean = clean[:, :, :ori_h, :ori_w] return clean class SCUNetPipeline(Pipeline): def __init__(self, scunet: SCUNet, cldm: ControlLDM, diffusion: Diffusion, cond_fn: Optional[Guidance], device: str) -> None: super().__init__(scunet, cldm, diffusion, cond_fn, device) @count_vram_usage def run_stage1(self, lq: torch.Tensor) -> torch.Tensor: if lq.shape[-1] > 1500: clean = [] batch_lq = lq.split(2, dim=0) for lq_ in batch_lq: clean.append(self.stage1_model(lq_)) torch.cuda.empty_cache() clean = torch.cat(clean) else: clean = self.stage1_model(lq) if min(clean.shape[2:]) < 512: clean = resize_short_edge_to(clean, size=512) # import ipdb; ipdb.set_trace() return clean