v0.1

Author: lizekai-richard

configs/Demo_Hard_Label.yaml

@@ -1,34 +1,36 @@
 
 # real data
-dataset: "CIFAR10"
-real_data_path: "./dataset/"
-custom_val_trans: None
+dataset: CIFAR10
+real_data_path: ./dataset/
+custom_val_trans: null
 
 # synthetic data
 ipc: 10
-im_size: (32, 32)
+im_size: [32, 32]
 
 # agent model
-model_name: "ConvNet-3"
-use_torchvision: False
+model_name: ConvNet-3
+use_torchvision: false
 
 # data augmentation
 data_aug_func: "dsa"
-aug_params: {
-  "prob_flip": 0.5, 
-  "ratio_rotate": 15.0, 
-  "saturation": 2.0, 
-  "brightness": 1.0, 
-  "contrast": 0.5, 
-  "ratio_scale": 1.2, 
-  "ratio_crop_pad": 0.125, 
-  "ratio_cutout": 0.5
-}
-use_zca: False
+aug_params:
+  prob_flip: 0.5
+  ratio_rotate: 15.0
+  saturation: 2.0
+  brightness: 1.0
+  contrast: 0.5
+  ratio_scale: 1.2
+  ratio_crop_pad: 0.125
+  ratio_cutout: 0.5
+use_zca: false
+
+custom_train_trans: null
+custom_val_trans: null
 
 # training specifics
-optimizer: "sgd"
-lr_scheduler: "step"
+optimizer: sgd
+lr_scheduler: step
 weight_decay: 0.0005
 momentum: 0.9
 num_eval: 5
@@ -37,7 +39,7 @@ syn_batch_size: 128
 real_batch_size: 256
 default_lr: 0.01
 num_workers: 4
-device: "cuda"
+device: cuda
 
 # save path
-save_path: "./results/my_method_hard_label_scores.csv"
+save_path: ./results/my_method_hard_label_scores.csv

configs/Demo_Soft_Label.yaml

@@ -1,49 +1,49 @@
 # real data
-dataset: "CIFAR10"
-real_data_path: "./dataset/"
-custom_val_trans: None
+dataset: CIFAR10
+real_data_path: ./dataset/
+custom_val_trans: null
 
 # synthetic data
 ipc: 10
-im_size: (32, 32)
+im_size: [32, 32]
 
 # agent model
-model_name: "ConvNet-3"
-stu_use_torchvision: False
-tea_use_torchvision: False
-teacher_dir: "./teacher_models"
+model_name: ConvNet-3
+stu_use_torchvision: false
+tea_use_torchvision: false
+teacher_dir: ./teacher_models
 
-# data augmentation
-data_aug_func: "dsa"
-aug_params: {
-    "prob_flip": 0.5,
-    "ratio_rotate": 15.0,
-    "saturation": 2.0,
-    "brightness": 1.0,
-    "contrast": 0.5,
-    "ratio_scale": 1.2,
-    "ratio_crop_pad": 0.125,
-    "ratio_cutout": 0.5
-}
-use_zca: True
+# syntheticdata augmentation
+data_aug_func: dsa
+aug_params:
+    prob_flip: 0.5
+    ratio_rotate: 15.0
+    saturation: 2.0
+    brightness: 1.0
+    contrast: 0.5
+    ratio_scale: 1.2
+    ratio_crop_pad: 0.125
+    ratio_cutout: 0.5
+use_zca: true
+custom_train_trans: null
 
 # soft label settings
-soft_label_mode: "S"
-soft_label_criterion: "sce"
+soft_label_mode: S
+soft_label_criterion: sce
 temperature: 1.0
 
 # training specifics
-optimizer: "sgd"
-lr_scheduler: "step"
+optimizer: sgd
+lr_scheduler: step
 weight_decay: 0.0005
 momentum: 0.9
 num_eval: 5
 num_epochs: 1000
 default_lr: 0.01
 num_workers: 4
-device: "cuda"
+device: cuda
 syn_batch_size: 128
 real_batch_size: 256
 
 # save path
-save_path: "./results/my_method_soft_label_scores.csv"
+save_path: ./my_method_soft_label_scores.csv

dd_ranking/config/user_config.py

@@ -1,7 +1,7 @@
 import yaml
 import json
 from typing import Dict, Any
-
+from torchvision import transforms
 
 class Config:
     """Configuration object to manage individual configurations."""
@@ -21,9 +21,30 @@ def from_file(cls, filepath: str):
         else:
             raise ValueError("Unsupported file format. Use YAML or JSON.")
         return cls(config)
+    
+    def load_transforms_from_yaml(self, values):
+        if values is None:
+            return None
+        transform_list = [] 
+        for transform in values:
+            name = transform["name"]
+            args = transform.get("args", [])
+            if isinstance(args, dict):
+                transform_list.append(getattr(transforms, name)(**args))
+            else:
+                transform_list.append(getattr(transforms, name)(*args))
+        
+        return transforms.Compose(transform_list)
 
     def get(self, key: str, default: Any = None):
         """Get a value from the config."""
+        if key == "custom_train_trans":
+            return self.load_transforms_from_yaml(self.config["custom_train_trans"])
+        elif key == "custom_val_trans":
+            return self.load_transforms_from_yaml(self.config["custom_val_trans"])
+        elif key == "im_size":
+            return tuple(self.config.get("im_size", default))
+        
         return self.config.get(key, default)
 
     def update(self, overrides: Dict[str, Any]):

dd_ranking/metrics/__init__.py

@@ -1,3 +1,3 @@
-from .dd_ranking_unified import Unified_Evaluator
-from .dd_ranking_obj import Soft_Label_Evaluator, Hard_Label_Evaluator
-# from .dd_ranking_aug import Augmentation_Evaluator, DSA_Augmentation_Evaluator, ZCA_Whitening_Augmentation_Evaluator, Mixup_Augmentation_Evaluator, Cutmix_Augmentation_Evaluator
+from .general import Unified_Evaluator
+from .soft_label import Soft_Label_Evaluator
+from .hard_label import Hard_Label_Evaluator

dd_ranking/metrics/general.py

@@ -207,7 +207,7 @@ def compute_metrics_helper(self, model, loader, lr):
         scheduler = get_lr_scheduler(optimizer, self.lr_scheduler, self.num_epochs)
 
         best_acc = 0
-        for epoch in range(self.num_epochs):
+        for epoch in tqdm(range(self.num_epochs), total=self.num_epochs, desc="Training"):
             train_one_epoch(
                 model=model, 
                 loader=loader, 

demo_hard.py

@@ -6,9 +6,12 @@
 
 """ Use config file to specify the arguments (Recommended) """
 config = Config.from_file("./configs/Demo_Hard_Label.yaml")
-convd3_hard_obj = Hard_Label_Objective_Metrics(config)
-syn_images = torch.load(os.path.join("./DC/CIFAR10/IPC10/", f"images.pt"), map_location='cpu')
-print(convd3_hard_obj.compute_metrics(syn_images, syn_lr=0.01))
+hard_label_evaluator = Hard_Label_Objective_Metrics(config)
+
+syn_data_dir = "./baselines/DM/CIFAR10/IPC10/"
+syn_images = torch.load(os.path.join(syn_data_dir, f"images.pt"), map_location='cpu')
+syn_lr = 0.01
+print(hard_label_evaluator.compute_metrics(syn_images, syn_lr=syn_lr))
 
 
 """ Use keyword arguments """
@@ -34,7 +37,7 @@
 
 syn_images = torch.load(os.path.join(syn_data_dir, f"images.pt"), map_location='cpu')
 save_path = f"./results/{dataset}/{model_name}/IPC{ipc}/dm_hard_scores.csv"
-convd3_hard_obj = Hard_Label_Objective_Metrics(
+hard_label_evaluator = Hard_Label_Objective_Metrics(
     dataset=dataset, 
     real_data_path=data_dir, 
     ipc=ipc, 
@@ -58,4 +61,4 @@
     device=device,
     save_path=save_path
 )
-print(convd3_hard_obj.compute_metrics(syn_images, syn_lr=0.01))
+print(hard_label_evaluator.compute_metrics(syn_images, syn_lr=0.01))

demo_soft.py

@@ -1,16 +1,20 @@
 import os
 import torch
-from dd_ranking.metrics import Soft_Label_Objective_Metrics
+import warnings
+from dd_ranking.metrics import Soft_Label_Evaluator
 from dd_ranking.config import Config
+warnings.filterwarnings("ignore", category=FutureWarning)
 
 
 """ Use config file to specify the arguments (Recommended) """
 config = Config.from_file("./configs/Demo_Soft_Label.yaml")
-convd3_soft_obj = Soft_Label_Objective_Metrics(config)
+soft_label_evaluator = Soft_Label_Evaluator(config)
+
+syn_data_dir = "./baselines/DATM/CIFAR10/IPC10/"
 syn_images = torch.load(os.path.join(syn_data_dir, f"images.pt"), map_location='cpu')
 soft_labels = torch.load(os.path.join(syn_data_dir, f"labels.pt"), map_location='cpu')
 syn_lr = torch.load(os.path.join(syn_data_dir, f"lr.pt"), map_location='cpu')
-print(convd3_soft_obj.compute_metrics(syn_images, soft_labels, syn_lr=syn_lr))
+print(soft_label_evaluator.compute_metrics(image_tensor=syn_images, soft_labels=soft_labels, syn_lr=syn_lr))
 
 
 """ Use keyword arguments """
@@ -35,8 +39,9 @@
 
 syn_images = torch.load(os.path.join(syn_data_dir, f"images.pt"), map_location='cpu')
 soft_labels = torch.load(os.path.join(syn_data_dir, f"labels.pt"), map_location='cpu')
+syn_lr = torch.load(os.path.join(syn_data_dir, f"lr.pt"), map_location='cpu')
 save_path = f"./results/{dataset}/{model_name}/IPC{ipc}/dm_hard_scores.csv"
-convd3_hard_obj = Soft_Label_Objective_Metrics(
+soft_label_evaluator = Soft_Label_Evaluator(
     dataset=dataset,
     real_data_path=data_dir, 
     ipc=ipc, 
@@ -64,4 +69,4 @@
     device=device,
     save_path=save_path
 )
-print(convd3_hard_obj.compute_metrics(syn_images, soft_labels, syn_lr=0.01))
+print(soft_label_evaluator.compute_metrics(syn_images, soft_labels, syn_lr=syn_lr))