Update ICCAD.py

ICCAD.py

@@ -39,8 +39,9 @@ class ICCAD_Config(datasets.BuilderConfig):
     patchify: bool = False
     # transform
     use_resize: bool = True
-    img_size: int = 288
-    dist_type: str = 'edt'
+    img_size: int = 384
+    dist_type: str = 'cdt'
+    clean_connection: bool = True
     use_i_map: bool = True
     use_r_map: bool = True
     use_r_dist: bool = True
@@ -62,12 +63,15 @@ def _transform(self, img, max_pool=False):
         img = patchify(img, patch_size=(img_size, img_size, img.shape[-1]), step=img_size)
         img = img.reshape((-1, img_size, img_size, img.shape[-1])).transpose((0, 3, 1, 2))
     elif self.config.use_resize:
-        img = A.Resize(img_size, img_size, interpolation=cv2.INTER_AREA)(image=img)["image"]
-        img = img.transpose((2, 0, 1))
-    elif max_pool:
-        img = torch.from_numpy(img).unsqueeze(0)
-        img = F.adaptive_max_pool2d(img, img_size)
-        img = img.squeeze(0).numpy()
+        if max_pool:
+            img = torch.from_numpy(img).unsqueeze(0)
+            img = F.adaptive_max_pool2d(img, img_size)
+            img = img.squeeze(0).numpy()
+            img = img.transpose((2, 0, 1))
+        else:
+            img = A.Resize(img_size, img_size, interpolation=cv2.INTER_AREA)(image=img)["image"]
+            img = img.transpose((2, 0, 1))
+    else:
         img = img.transpose((2, 0, 1))
     
     return img
@@ -138,7 +142,8 @@ def get_image(self, data):
             get_dist(self, data["R_map_88"]) * get_blur(data["I_map"], ksize=49),
             get_dist(self, data["R_map_89"]) * get_blur(data["I_map"], ksize=49),
             get_dist(self, data["R_map_99"]) * get_blur(data["I_map"], ksize=49),
-        ] if self.config.use_power_map else []), axis=2))
+        ] if self.config.use_power_map else []), axis=2)),
+        "ir_drop": data["ir_drop"],
     }
     
 def get_IR_map_H(Via_map, I_map):
@@ -297,13 +302,23 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                 "H": datasets.Value("int32"),
                 "W": datasets.Value("int32"),
                 "image": datasets.Array4D((None, in_chans, img_size, img_size), 'float32'),
+                "ir_drop": datasets.Image(),
             })
-        else:
+        elif self.config.use_resize:
             features = datasets.Features({
                 "data_idx": datasets.Value("string"),
                 "H": datasets.Value("int32"),
                 "W": datasets.Value("int32"),
                 "image": datasets.Array3D((in_chans, img_size, img_size), 'float32'),
+                "ir_drop": datasets.Image(),
+            })
+        else:
+            features = datasets.Features({
+                "data_idx": datasets.Value("string"),
+                "H": datasets.Value("int32"),
+                "W": datasets.Value("int32"),
+                "image": datasets.Array3D((in_chans, 1, -1), 'float32'),
+                "ir_drop": datasets.Image(),
             })
 
         return datasets.DatasetInfo(
@@ -316,68 +331,45 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
         fake_cur = []
         fake_pdn = []
         fake_dist = []
-        fake_irdrop = []
+        fake_ir_drop = []
         fake_netlist = []
         
         real_idx = []
         real_cur = []
         real_pdn = []
         real_dist = []
-        real_irdrop = []
+        real_ir_drop = []
         real_netlist = []
         
-        if self.config.use_BeGAN:
-            BeGAN_01_idx = []
-            BeGAN_01_cur = []
-            BeGAN_01_pdn = []
-            BeGAN_01_dist = []
-            BeGAN_01_irdrop = []
-            BeGAN_01_netlist = []
-
-            BeGAN_02_idx = []
-            BeGAN_02_cur = []
-            BeGAN_02_pdn = []
-            BeGAN_02_dist = []
-            BeGAN_02_irdrop = []
-            BeGAN_02_netlist = []
-        
+        BeGAN_01_idx = []
+        BeGAN_01_cur = []
+        BeGAN_01_pdn = []
+        BeGAN_01_dist = []
+        BeGAN_01_ir_drop = []
+        BeGAN_01_netlist = []
+
+        BeGAN_02_idx = []
+        BeGAN_02_cur = []
+        BeGAN_02_pdn = []
+        BeGAN_02_dist = []
+        BeGAN_02_ir_drop = []
+        BeGAN_02_netlist = []
         
         # Download images
-        fake_data_files = os.path.join(dl_manager.download_and_extract(_URLS["fake_data_url"]), "fake-circuit-data_20230623")
         real_data_files = os.path.join(dl_manager.download_and_extract(_URLS["real_data_url"]), "real-circuit-data_20230615")
-
-        fake_path_files = sorted(glob.glob(os.path.join(fake_data_files, "*.sp")))
         real_path_files = sorted(glob.glob(os.path.join(real_data_files, "*")))
         
-        if self.config.use_BeGAN:
+        if not self.config.test_mode:
+            fake_data_files = os.path.join(dl_manager.download_and_extract(_URLS["fake_data_url"]), "fake-circuit-data_20230623")
+            fake_path_files = sorted(glob.glob(os.path.join(fake_data_files, "*.sp")))
+        
+        if self.config.use_BeGAN and not self.config.test_mode:
             BeGAN_01_data_files = os.path.join(dl_manager.download_and_extract(_URLS["BeGAN_01_data_url"]), "BeGAN-ver01")
             BeGAN_01_path_files = sorted(glob.glob(os.path.join(BeGAN_01_data_files, "*.sp")))
 
             BeGAN_02_data_files = os.path.join(dl_manager.download_and_extract(_URLS["BeGAN_02_data_url"]), "BeGAN-ver02")
             BeGAN_02_path_files = sorted(glob.glob(os.path.join(BeGAN_02_data_files, "*.sp")))
         
-        # for fake
-        for path in fake_path_files:
-            data_idx = os.path.basename(path).split(".")[0]
-            fake_idx.append(data_idx)
-            data_path = glob.glob(os.path.join(os.path.dirname(path), data_idx + "*.*"))
-            
-            for data in data_path:
-                if "current.csv" in os.path.basename(data):
-                    fake_cur.append(data)
-                elif "eff_dist.csv" in os.path.basename(data):
-                    fake_dist.append(data)
-                elif "ir_drop.csv" in os.path.basename(data):
-                    fake_irdrop.append(data)
-                elif "pdn_density.csv" in os.path.basename(data):
-                    fake_pdn.append(data)
-                elif ".sp" in os.path.basename(data):
-                    fake_netlist.append(data)
-                else:
-                    raise AssertionError(os.path.basename(data), "fake data path error")
-            
-        assert len(fake_idx) == len(fake_cur) == len(fake_dist) == len(fake_irdrop) == len(fake_pdn) == len(fake_netlist), f"{(len(fake_idx), len(fake_cur), len(fake_dist), len(fake_irdrop), len(fake_pdn), len(fake_netlist))} fake data length not the same"
-        
         # for real
         for path in real_path_files:
             data_idx = os.path.basename(path)
@@ -390,16 +382,40 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                 elif "eff_dist_map.csv" in os.path.basename(data):
                     real_dist.append(data)
                 elif "ir_drop_map.csv" in os.path.basename(data):
-                    real_irdrop.append(data)
+                    real_ir_drop.append(data)
                 elif "pdn_density.csv" in os.path.basename(data):
                     real_pdn.append(data)
                 elif "netlist.sp" in os.path.basename(data):
                     real_netlist.append(data)
                 else:
                     raise AssertionError(os.path.basename(data), "real data path error")
-                    
-        assert len(real_idx) == len(real_cur) == len(real_dist) == len(real_irdrop) == len(real_pdn) == len(real_netlist), f"{(len(real_idx), len(real_cur), len(real_dist), len(real_irdrop), len(real_pdn), len(real_netlist))} real data length not the same"
-        if self.config.use_BeGAN:
+                
+            assert len(real_idx) == len(real_cur) == len(real_dist) == len(real_ir_drop) == len(real_pdn) == len(real_netlist), f"{(len(real_idx), len(real_cur), len(real_dist), len(real_ir_drop), len(real_pdn), len(real_netlist))} real data length not the same"
+        
+        # for fake
+        if not self.config.test_mode:
+            for path in fake_path_files:
+                data_idx = os.path.basename(path).split(".")[0]
+                fake_idx.append(data_idx)
+                data_path = glob.glob(os.path.join(os.path.dirname(path), data_idx + "*.*"))
+
+                for data in data_path:
+                    if "current.csv" in os.path.basename(data):
+                        fake_cur.append(data)
+                    elif "eff_dist.csv" in os.path.basename(data):
+                        fake_dist.append(data)
+                    elif "ir_drop.csv" in os.path.basename(data):
+                        fake_ir_drop.append(data)
+                    elif "pdn_density.csv" in os.path.basename(data):
+                        fake_pdn.append(data)
+                    elif ".sp" in os.path.basename(data):
+                        fake_netlist.append(data)
+                    else:
+                        raise AssertionError(os.path.basename(data), "fake data path error")
+
+            assert len(fake_idx) == len(fake_cur) == len(fake_dist) == len(fake_ir_drop) == len(fake_pdn) == len(fake_netlist), f"{(len(fake_idx), len(fake_cur), len(fake_dist), len(fake_ir_drop), len(fake_pdn), len(fake_netlist))} fake data length not the same"
+
+        if self.config.use_BeGAN and not self.config.test_mode:
             # for BeGAN-ver01
             for path in BeGAN_01_path_files:
                 data_idx = os.path.basename(path).split(".")[0]
@@ -412,7 +428,7 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                     elif "eff_dist.csv" in os.path.basename(data):
                         BeGAN_01_dist.append(data)
                     elif "ir_drop_map.csv" in os.path.basename(data):
-                        BeGAN_01_irdrop.append(data)
+                        BeGAN_01_ir_drop.append(data)
                     elif "pdn_density.csv" in os.path.basename(data):
                         BeGAN_01_pdn.append(data)
                     elif ".sp" in os.path.basename(data):
@@ -420,7 +436,7 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                     else:
                         raise AssertionError(os.path.basename(data), "BeGAN-ver01 data path error")
 
-            assert len(BeGAN_01_idx) == len(BeGAN_01_cur) == len(BeGAN_01_dist) == len(BeGAN_01_irdrop) == len(BeGAN_01_pdn) == len(BeGAN_01_netlist), f"{(len(BeGAN_01_idx), len(BeGAN_01_cur), len(BeGAN_01_dist), len(BeGAN_01_irdrop), len(BeGAN_01_pdn), len(BeGAN_01_netlist))} BeGAN-ver02 data length not the same"
+            assert len(BeGAN_01_idx) == len(BeGAN_01_cur) == len(BeGAN_01_dist) == len(BeGAN_01_ir_drop) == len(BeGAN_01_pdn) == len(BeGAN_01_netlist), f"{(len(BeGAN_01_idx), len(BeGAN_01_cur), len(BeGAN_01_dist), len(BeGAN_01_ir_drop), len(BeGAN_01_pdn), len(BeGAN_01_netlist))} BeGAN-ver02 data length not the same"
 
             # for BeGAN-ver02
             for path in BeGAN_02_path_files:
@@ -434,7 +450,7 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                     elif "eff_dist.csv" in os.path.basename(data):
                         BeGAN_02_dist.append(data)
                     elif "voltage.csv" in os.path.basename(data):
-                        BeGAN_02_irdrop.append(data)
+                        BeGAN_02_ir_drop.append(data)
                     elif "regions.csv" in os.path.basename(data):
                         BeGAN_02_pdn.append(data)
                     elif ".sp" in os.path.basename(data):
@@ -442,60 +458,61 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                     else:
                         raise AssertionError(os.path.basename(data), "BeGAN-ver01 data path error")
 
-            assert len(BeGAN_02_idx) == len(BeGAN_02_cur) == len(BeGAN_02_dist) == len(BeGAN_02_irdrop) == len(BeGAN_02_pdn) == len(BeGAN_02_netlist), f"{(len(BeGAN_02_idx), len(BeGAN_02_cur), len(BeGAN_02_dist), len(BeGAN_02_irdrop), len(BeGAN_02_pdn), len(BeGAN_02_netlist))} BeGAN-ver01 data length not the same"
-            
-        return [datasets.SplitGenerator(
-                name=datasets.Split("real"),
-                gen_kwargs={
-                    "data_idx": real_idx,
-                    "current": real_cur,
-                    "pdn_density": real_pdn,
-                    "eff_dist": real_dist,
-                    "ir_drop": real_irdrop,
-                    "netlist": real_netlist,
-                })
-            ] if self.config.test_mode else [
-                datasets.SplitGenerator(
-                name=datasets.Split("fake"),
-                gen_kwargs={
-                    "data_idx": fake_idx,
-                    "current": fake_cur,
-                    "pdn_density": fake_pdn,
-                    "eff_dist": fake_dist,
-                    "ir_drop": fake_irdrop,
-                    "netlist": fake_netlist,
-                })
-            ] + [datasets.SplitGenerator(
-                name=datasets.Split("real"),
-                gen_kwargs={
-                    "data_idx": real_idx,
-                    "current": real_cur,
-                    "pdn_density": real_pdn,
-                    "eff_dist": real_dist,
-                    "ir_drop": real_irdrop,
-                    "netlist": real_netlist,
-                })
-            ] + ([datasets.SplitGenerator(
-                name=datasets.Split("BeGAN_01"),
-                gen_kwargs={
-                    "data_idx": BeGAN_01_idx,
-                    "current": BeGAN_01_cur,
-                    "pdn_density": BeGAN_01_pdn,
-                    "eff_dist": BeGAN_01_dist,
-                    "ir_drop": BeGAN_01_irdrop,
-                    "netlist": BeGAN_01_netlist,
-                }),
-                datasets.SplitGenerator(
-                name=datasets.Split("BeGAN_02"),
-                gen_kwargs={
-                    "data_idx": BeGAN_02_idx,
-                    "current": BeGAN_02_cur,
-                    "pdn_density": BeGAN_02_pdn,
-                    "eff_dist": BeGAN_02_dist,
-                    "ir_drop": BeGAN_02_irdrop,
-                    "netlist": BeGAN_02_netlist,
-                })
-            ] if self.config.use_BeGAN else [])
+            assert len(BeGAN_02_idx) == len(BeGAN_02_cur) == len(BeGAN_02_dist) == len(BeGAN_02_ir_drop) == len(BeGAN_02_pdn) == len(BeGAN_02_netlist), f"{(len(BeGAN_02_idx), len(BeGAN_02_cur), len(BeGAN_02_dist), len(BeGAN_02_ir_drop), len(BeGAN_02_pdn), len(BeGAN_02_netlist))} BeGAN-ver01 data length not the same"
+        
+        if self.config.test_mode:
+            return [datasets.SplitGenerator(
+                    name=datasets.Split("real"),
+                    gen_kwargs={
+                        "data_idx": real_idx,
+                        "current": real_cur,
+                        "pdn_density": real_pdn,
+                        "eff_dist": real_dist,
+                        "ir_drop": real_ir_drop,
+                        "netlist": real_netlist,
+                    })]
+        else:
+            return [datasets.SplitGenerator(
+                    name=datasets.Split("fake"),
+                    gen_kwargs={
+                        "data_idx": fake_idx,
+                        "current": fake_cur,
+                        "pdn_density": fake_pdn,
+                        "eff_dist": fake_dist,
+                        "ir_drop": fake_ir_drop,
+                        "netlist": fake_netlist,
+                    })
+                ] + [datasets.SplitGenerator(
+                    name=datasets.Split("real"),
+                    gen_kwargs={
+                        "data_idx": real_idx,
+                        "current": real_cur,
+                        "pdn_density": real_pdn,
+                        "eff_dist": real_dist,
+                        "ir_drop": real_ir_drop,
+                        "netlist": real_netlist,
+                    })
+                ] + ([datasets.SplitGenerator(
+                    name=datasets.Split("BeGAN_01"),
+                    gen_kwargs={
+                        "data_idx": BeGAN_01_idx,
+                        "current": BeGAN_01_cur,
+                        "pdn_density": BeGAN_01_pdn,
+                        "eff_dist": BeGAN_01_dist,
+                        "ir_drop": BeGAN_01_ir_drop,
+                        "netlist": BeGAN_01_netlist,
+                    }),
+                    datasets.SplitGenerator(
+                    name=datasets.Split("BeGAN_02"),
+                    gen_kwargs={
+                        "data_idx": BeGAN_02_idx,
+                        "current": BeGAN_02_cur,
+                        "pdn_density": BeGAN_02_pdn,
+                        "eff_dist": BeGAN_02_dist,
+                        "ir_drop": BeGAN_02_ir_drop,
+                        "netlist": BeGAN_02_netlist,
+                    })
+                ] if self.config.use_BeGAN else [])
       
       
     def _generate_examples(self, data_idx, current, pdn_density, eff_dist, ir_drop, netlist):
@@ -761,16 +778,16 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
                 else:
                     raise AssertionError(_data_idx, "R map layer not found", row[0], row[3])
             
-            # clean not connect
-            # R_map_99[:, ~V_map.any(axis=0)] = 0
-            # R_map_89[:, ~V_map.any(axis=0)] = 0
-            
             # clean wrong via
             R_map_14[(R_map_11 == 0) | (R_map_44 == 0)] = 0
             R_map_47[(R_map_44 == 0) | (R_map_77 == 0)] = 0
             R_map_78[(R_map_77 == 0) | (R_map_88 == 0)] = 0
             R_map_89[(R_map_88 == 0) | (R_map_99 == 0)] = 0
-
+            
+            # clean not connect
+            if self.config.clean_connection:
+                R_map_99[:, ~V_map.any(axis=0)] = 0
+                R_map_89[:, ~V_map.any(axis=0)] = 0
             
             # I
             df_I = df[df["type"].str.contains("I")]
@@ -830,29 +847,30 @@ class ICCAD_Dataset(datasets.GeneratorBasedBuilder):
 
 
             # resize
-            R_map_11 = resize(R_map_11) / (H * W) * (_H * _W)
-            R_map_14 = resize(R_map_14) / (H * W) * (_H * _W)
-            R_map_44 = resize(R_map_44) / (H * W) * (_H * _W)
-            R_map_47 = resize(R_map_47) / (H * W) * (_H * _W)
-            R_map_77 = resize(R_map_77) / (H * W) * (_H * _W)
-            R_map_78 = resize(R_map_78) / (H * W) * (_H * _W)
-            R_map_88 = resize(R_map_88) / (H * W) * (_H * _W)
-            R_map_89 = resize(R_map_89) / (H * W) * (_H * _W)
-            R_map_99 = resize(R_map_99) / (H * W) * (_H * _W)
-            I_map = resize(I_map) / (H * W) * (_H * _W)
-            V_map = resize(V_map) / (H * W) * (_H * _W)
+            R_map_11 = resize(R_map_11) * (H * W) / (_H * _W)
+            R_map_14 = resize(R_map_14) * (H * W) / (_H * _W)
+            R_map_44 = resize(R_map_44) * (H * W) / (_H * _W)
+            R_map_47 = resize(R_map_47) * (H * W) / (_H * _W)
+            R_map_77 = resize(R_map_77) * (H * W) / (_H * _W)
+            R_map_78 = resize(R_map_78) * (H * W) / (_H * _W)
+            R_map_88 = resize(R_map_88) * (H * W) / (_H * _W)
+            R_map_89 = resize(R_map_89) * (H * W) / (_H * _W)
+            R_map_99 = resize(R_map_99) * (H * W) / (_H * _W)
+            I_map = resize(I_map) * (H * W) / (_H * _W)
+            V_map = resize(V_map) * (H * W) / (_H * _W)
             if self.config.use_multi_dist:
-                V_multi_dist = resize(1 / V_multi_dist) / math.sqrt(H * W) * math.sqrt(_H * _W)
-                R_multi_dist_14 = resize(1 / R_multi_dist_14) / math.sqrt(H * W) * math.sqrt(_H * _W)
-                R_multi_dist_47 = resize(1 / R_multi_dist_47) / math.sqrt(H * W) * math.sqrt(_H * _W)
-                R_multi_dist_78 = resize(1 / R_multi_dist_78) / math.sqrt(H * W) * math.sqrt(_H * _W)
-                R_multi_dist_89 = resize(1 / R_multi_dist_89) / math.sqrt(H * W) * math.sqrt(_H * _W)
+                V_multi_dist = resize(1 / V_multi_dist) * math.sqrt(H * W) / math.sqrt(_H * _W)
+                R_multi_dist_14 = resize(1 / R_multi_dist_14) * math.sqrt(H * W) / math.sqrt(_H * _W)
+                R_multi_dist_47 = resize(1 / R_multi_dist_47) * math.sqrt(H * W) / math.sqrt(_H * _W)
+                R_multi_dist_78 = resize(1 / R_multi_dist_78) * math.sqrt(H * W) / math.sqrt(_H * _W)
+                R_multi_dist_89 = resize(1 / R_multi_dist_89) * math.sqrt(H * W) / math.sqrt(_H * _W)
             if self.config.use_ir_map:
-                IR_map_14 = resize(IR_map_14, max_pool=False) / (H * W) * (_H * _W) / (H * W) * (_H * _W)
-                IR_map_47 = resize(IR_map_47, max_pool=False) / (H * W) * (_H * _W) / (H * W) * (_H * _W)
-                IR_map_78 = resize(IR_map_78, max_pool=False) / (H * W) * (_H * _W) / (H * W) * (_H * _W)
-                IR_map_89 = resize(IR_map_89, max_pool=False) / (H * W) * (_H * _W) / (H * W) * (_H * _W)
+                IR_map_14 = resize(IR_map_14, max_pool=False)
+                IR_map_47 = resize(IR_map_47, max_pool=False)
+                IR_map_78 = resize(IR_map_78, max_pool=False)
+                IR_map_89 = resize(IR_map_89, max_pool=False)
 
+            img_size = self.config.img_size
             
             yield _idx, get_image(self, {
                 **{