Update ForNet.py

add fg_in_nonant and fg_size_fact

ForNet.py

@@ -1073,6 +1073,8 @@ class RecombineDataset(Dataset):
         mask_smoothing_sigma,
         rel_jut_out,
         orig_img_prob,
+        fg_in_nonant=None,
+        size_fact=1.0,
         **kwargs,
     ):
         """Create the ForNet recombination dataset.
@@ -1099,6 +1101,7 @@ class RecombineDataset(Dataset):
             "orig",
         ], f"Invalid background_combination {background_combination}"
         assert fg_size_mode in ["range", "min", "max", "mean"], f"Invalid fg_size_mode {fg_size_mode}"
+        assert fg_in_nonant is None or -1 <= fg_in_nonant < 9, f"fg_in_nonant={fg_in_nonant} not in [0, 8] or None"
         self.background_combination = background_combination
         self.fg_scale_jitter = fg_scale_jitter
         self.pruning_ratio = pruning_ratio
@@ -1110,6 +1113,8 @@ class RecombineDataset(Dataset):
         self.epochs = 0
         self._epoch = 0
         self.cls_to_idx = {}
+        self.fg_in_nonant = fg_in_nonant
+        self.size_fact = size_fact
 
         bg_rat_indices = super()._getitem(0)["bg_rat_idx_file"]
         self.train = "train" in bg_rat_indices.split("/")[-1]
@@ -1177,6 +1182,9 @@ class RecombineDataset(Dataset):
         bg_img = bg_item["bg"].convert("RGB")
         bg_size = bg_img.size
         bg_area = bg_size[0] * bg_size[1]
+        if self.fg_in_nonant is not None:
+            bg_area = bg_area / 9
+
         orig_fg_ratio = fg_item["fg/bg_area"]
         bg_fg_ratio = bg_item["fg/bg_area"]
 
@@ -1198,6 +1206,7 @@ class RecombineDataset(Dataset):
                 goal_fg_ratio_lower * (1 - self.fg_scale_jitter), goal_fg_ratio_upper * (1 + self.fg_scale_jitter)
             )
             / fg_size_factor
+            * self.size_fact
         )
 
         goal_shape_y = round(np.sqrt(bg_area * fg_scale * fg_img.size[1] / fg_img.size[0]))
@@ -1228,6 +1237,12 @@ class RecombineDataset(Dataset):
         y_min = -self.rel_jut_out * fg_img.size[1]
         y_max = bg_size[1] - fg_img.size[1] * (1 - self.rel_jut_out)
 
+        if self.fg_in_nonant is not None and self.fg_in_nonant >= 0:
+            x_min = (self.fg_in_nonant % 3) * bg_size[0] / 3
+            x_max = ((self.fg_in_nonant % 3) + 1) * bg_size[0] / 3 - fg_img.size[0]
+            y_min = (self.fg_in_nonant // 3) * bg_size[1] / 3
+            y_max = ((self.fg_in_nonant // 3) + 1) * bg_size[1] / 3 - fg_img.size[1]
+
         if x_min > x_max:
             x_min = x_max = (x_min + x_max) / 2
         if y_min > y_max:
@@ -1262,6 +1277,8 @@ _CONFIG_HASH_IGNORE_KWARGS = [
     "mask_smoothing_sigma",
     "rel_jut_out",
     "orig_img_prob",
+    "fg_in_nonant",
+    "size_fact",
 ]
 
 
@@ -1278,6 +1295,8 @@ class ForNetConfig(datasets.BuilderConfig):
         mask_smoothing_sigma,
         rel_jut_out,
         orig_img_prob,
+        fg_in_nonant=None,
+        size_fact=1.0,
         **kwargs,
     ):
         """BuilderConfig for ForNet.
@@ -1294,6 +1313,8 @@ class ForNetConfig(datasets.BuilderConfig):
         self.mask_smoothing_sigma = mask_smoothing_sigma
         self.rel_jut_out = rel_jut_out
         self.orig_img_prob = orig_img_prob
+        self.fg_in_nonant = fg_in_nonant
+        self.size_fact = size_fact
 
     def __str__(self):
         return f"ForNetConfig(name={self.name}, version={self.version}, data_dir={self.data_dir}, data_files={self.data_files}, description={self.description}, background_combination={self.background_combination}, fg_scale_jitter={self.fg_scale_jitter}, pruning_ratio={self.pruning_ratio}, fg_size_mode={self.fg_size_mode}, fg_bates_n={self.fg_bates_n}, mask_smoothing_sigma={self.mask_smoothing_sigma}, rel_jut_out={self.rel_jut_out}, orig_img_prob={self.orig_img_prob})"
@@ -1608,6 +1629,8 @@ class ForNet(datasets.GeneratorBasedBuilder):
             mask_smoothing_sigma=self.config.mask_smoothing_sigma,
             rel_jut_out=self.config.rel_jut_out,
             orig_img_prob=self.config.orig_img_prob,
+            fg_in_nonant=self.config.fg_in_nonant,
+            size_fact=self.config.size_fact,
             **dataset_kwargs,
         )