ootd

.gitignore

@@ -10,3 +10,4 @@ train.txt
 VITON*
 eval_output
 eval_ootd.py
+metrics_aigc
\ No newline at end of file

README.md

@@ -164,7 +164,11 @@ https://hf-mirror.com/openai/clip-vit-large-patch14/tree/main
 
 ### 精度
 
-待补充
+|ssim|lpips|
+|:---:|:---:|
+|0.86|0.075|
+
+注意：该精度在size=(512, 384)条件下训练及测试得到，与官方实现（未开源）可能存在不同。
 
 ## 应用场景
 

ootd/inference_ootd_hd.py

@@ -30,8 +30,8 @@ sys.path.append(str(OOTD_ROOT))
 # VIT_PATH = "../checkpoints/clip-vit-large-patch14"
 VIT_PATH = os.path.join(OOTD_ROOT, "checkpoints/clip-vit-large-patch14")
 VAE_PATH = "../checkpoints/ootd"
-UNET_PATH = "../checkpoints/ootd/ootd_hd/checkpoint-36000"
-# UNET_PATH = "../train/checkpoints"
+# UNET_PATH = "../checkpoints/ootd/ootd_hd/checkpoint-36000"
+UNET_PATH = "../train/ckpts_bak/checkpoints"
 MODEL_PATH = "../checkpoints/ootd"
 
 class OOTDiffusionHD:
@@ -123,6 +123,8 @@ class OOTDiffusionHD:
             else:
                 raise ValueError("model_type must be \'hd\' or \'dc\'!")
 
+            # start = time.time()
+            # with torch.profiler.profile(activities=[torch.profiler.ProfilerActivity.CUDA]) as p:
             images = self.pipe(prompt_embeds=prompt_embeds,
                         image_garm=image_garm,
                         image_vton=image_vton, 
@@ -133,5 +135,11 @@ class OOTDiffusionHD:
                         num_images_per_prompt=num_samples,
                         generator=generator,
             ).images
+            # print(p.key_averages().table(sort_by="self_cuda_time_total", row_limit=-1))
+            # p.export_chrome_trace("trace.json")
+            
+            # end = time.time()
+            
+            # print(f"Inference time: {end - start} s")
 
         return images

train/data/viton.py

@@ -45,16 +45,19 @@ class VITONDataset(data.Dataset):
         self.c_names['paired'] = img_names
 
     def get_parse_agnostic(self, parse, pose_data):  
+        # parse 语义分割图
+        # pose_data 姿势信息
         parse_array = np.array(parse)
         parse_upper = ((parse_array == 5).astype(np.float32) +
                        (parse_array == 6).astype(np.float32) +
-                       (parse_array == 7).astype(np.float32))
+                       (parse_array == 7).astype(np.float32))  # 这里是什么形式，应该是一张图且图中仅有这些部位
         parse_neck = (parse_array == 10).astype(np.float32)
 
         r = 10
         agnostic = parse.copy()
 
         # mask arms
+        # 14表示左臂，15表示右臂
         for parse_id, pose_ids in [(14, [2, 5, 6, 7]), (15, [5, 2, 3, 4])]:
             mask_arm = Image.new('L', (self.load_width, self.load_height), 'black')
             mask_arm_draw = ImageDraw.Draw(mask_arm)
@@ -129,18 +132,18 @@ class VITONDataset(data.Dataset):
     def __getitem__(self, index):
         img_name = self.img_names[index]
         c_name = {}
-        c = {}
-        cm = {}
+        c = {}  # 衣物
+        cm = {}  # 衣物的mask
         for key in self.c_names:
             c_name[key] = self.c_names[key][index]
-            c[key] = Image.open(osp.join(self.data_path, 'cloth', c_name[key])).convert('RGB')
-            c[key] = transforms.Resize(self.load_width, interpolation=2)(c[key])
+            c[key] = Image.open(osp.join(self.data_path, 'cloth', c_name[key])).convert('RGB')  # 读取衣服图像
+            c[key] = transforms.Resize(self.load_width, interpolation=2)(c[key])  # 修改宽度
             cm[key] = Image.open(osp.join(self.data_path, 'cloth-mask', c_name[key]))
             cm[key] = transforms.Resize(self.load_width, interpolation=0)(cm[key])
 
             c[key] = self.transform(c[key])  # [-1,1]
             cm_array = np.array(cm[key])
-            cm_array = (cm_array >= 128).astype(np.float32)
+            cm_array = (cm_array >= 128).astype(np.float32)  # 二值化
             cm[key] = torch.from_numpy(cm_array)  # [0,1]
             cm[key].unsqueeze_(0)
 
@@ -157,7 +160,7 @@ class VITONDataset(data.Dataset):
             pose_data = np.array(pose_data)
             pose_data = pose_data.reshape((-1, 3))[:, :2]
 
-        # load parsing image
+        # load parsing image  语义分割图
         parse_name = img_name.replace('.jpg', '.png')
         parse = Image.open(osp.join(self.data_path, 'image-parse-v3', parse_name))
         parse = transforms.Resize(self.load_width, interpolation=0)(parse)
@@ -179,6 +182,7 @@ class VITONDataset(data.Dataset):
             11: ['socks', [8]],
             12: ['noise', [3, 11]]
         }
+        # 不同通道表示不同类别
         parse_agnostic_map = torch.zeros(20, self.load_height, self.load_width, dtype=torch.float)
         parse_agnostic_map.scatter_(0, parse_agnostic, 1.0)
         new_parse_agnostic_map = torch.zeros(self.semantic_nc, self.load_height, self.load_width, dtype=torch.float)

train/main.py

@@ -59,15 +59,15 @@ def main():
                          args.lr_scheduler)
     
     trainer = L.Trainer(
-        max_epochs=50,
+        max_epochs=100,
         accelerator='auto',
         log_every_n_steps=1,
-        callbacks=[ModelCheckpoint(every_n_train_steps=6000, save_top_k=-1, save_last=True)],
+        callbacks=[ModelCheckpoint(every_n_train_steps=5000, save_top_k=-1, save_last=True)],
         precision="16-mixed",
         accumulate_grad_batches=32,
     )
     
-    trainer.fit(model, dm, ckpt_path="lightning_logs/version_6/checkpoints/last.ckpt")
+    trainer.fit(model, dm, ckpt_path="lightning_logs/version_11/checkpoints/epoch=54-step=10000.ckpt")
 
 
 if __name__ == "__main__":