birder 0.4.1__py3-none-any.whl → 0.4.2__py3-none-any.whl

birder/net/vit_sam.py

@@ -29,7 +29,9 @@ from birder.net._vit_configs import BASE
 from birder.net._vit_configs import HUGE
 from birder.net._vit_configs import LARGE
 from birder.net._vit_configs import MEDIUM
+from birder.net._vit_configs import SMALL
 from birder.net.base import DetectorBackbone
+from birder.net.base import normalize_out_indices
 from birder.net.vit import EncoderBlock as MAEDecoderBlock
 
 
@@ -72,7 +74,7 @@ def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor
 
     # Interpolate rel pos if needed
     if rel_pos.shape[0] != max_rel_dist:
-        # Adjust size is a one off interpolation, should prevent us from getting here
+        # Only reached in dynamic-size mode (rel-pos table resized on the fly)
         rel_pos_resized = F.interpolate(
             rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1), size=max_rel_dist, mode="linear"
         )
@@ -242,6 +244,7 @@ class EncoderBlock(nn.Module):
 class ViT_SAM(DetectorBackbone):
     block_group_regex = r"body\.(\d+)"
 
+    # pylint: disable=too-many-locals
     def __init__(
         self,
         input_channels: int,
@@ -266,6 +269,7 @@ class ViT_SAM(DetectorBackbone):
         window_size: int = self.config["window_size"]
         global_attn_indexes: list[int] = self.config["global_attn_indexes"]
         neck_channels: Optional[int] = self.config.get("neck_channels", None)
+        out_indices: Optional[list[int]] = self.config.get("out_indices", None)
         drop_path_rate: float = self.config["drop_path_rate"]
 
         if norm_layer_type == "LayerNorm":
@@ -292,6 +296,7 @@ class ViT_SAM(DetectorBackbone):
         self.hidden_dim = hidden_dim
         self.global_attn_indexes = global_attn_indexes
         self.num_special_tokens = 0
+        self.out_indices = normalize_out_indices(out_indices, num_layers)
         dpr = [x.item() for x in torch.linspace(0, drop_path_rate, num_layers)]  # Stochastic depth decay rule
 
         self.patch_embed = PatchEmbed(
@@ -356,8 +361,10 @@ class ViT_SAM(DetectorBackbone):
             nn.Flatten(1),
         )
 
-        self.return_stages = ["neck"]  # Actually meaningless, but for completeness
-        self.return_channels = [neck_channels]
+        num_return_stages = len(self.out_indices) if self.out_indices is not None else 1
+        self.return_stages = [f"stage{stage_idx + 1}" for stage_idx in range(num_return_stages)]
+        self.return_channels = [hidden_dim] * num_return_stages
+        self.return_channels[-1] = neck_channels
         self.embedding_size = neck_channels
         self.classifier = self.create_classifier()
 
@@ -372,13 +379,54 @@ class ViT_SAM(DetectorBackbone):
             activation_layer=nn.GELU,
         )
 
+    def _get_pos_embed(self, H: int, W: int) -> torch.Tensor:
+        if self.dynamic_size is False:
+            return self.pos_embedding
+
+        if H == self.size[0] and W == self.size[1]:
+            return self.pos_embedding
+
+        base_h = H // self.patch_size
+        base_w = W // self.patch_size
+        orig_dtype = self.pos_embedding.dtype
+        pos_embedding = self.pos_embedding.float()
+        pos_embedding = pos_embedding.permute(0, 3, 1, 2)
+        pos_embedding = F.interpolate(pos_embedding, size=(base_h, base_w), mode="bicubic", antialias=True)
+        pos_embedding = pos_embedding.permute(0, 2, 3, 1)
+
+        return pos_embedding.to(orig_dtype)
+
+    def set_causal_attention(self, is_causal: bool = True) -> None:
+        for b in self.body:
+            b.set_causal_attention(is_causal)
+
     def detection_features(self, x: torch.Tensor) -> dict[str, torch.Tensor]:
+        H, W = x.shape[-2:]
         x = self.patch_embed(x)
-        x = x + self.pos_embedding
+        x = x + self._get_pos_embed(H, W)
 
-        x = self.body(x)
-        x = self.neck(x.permute(0, 3, 1, 2))
-        return {self.return_stages[0]: x}
+        if self.out_indices is None:
+            x = self.body(x)
+            x = self.neck(x.permute(0, 3, 1, 2))
+            return {self.return_stages[0]: x}
+
+        out_indices_set = set(self.out_indices)
+        last_out_idx = max(out_indices_set)
+        out: dict[str, torch.Tensor] = {}
+        stage_idx = 0
+        for idx, blk in enumerate(self.body):
+            x = blk(x)
+            if idx not in out_indices_set:
+                continue
+
+            stage_x = x.permute(0, 3, 1, 2)
+            if idx == last_out_idx:
+                stage_x = self.neck(stage_x)
+
+            out[self.return_stages[stage_idx]] = stage_x
+            stage_idx += 1
+
+        return out
 
     def freeze_stages(self, up_to_stage: int) -> None:
         for param in self.patch_embed.parameters():
@@ -393,13 +441,10 @@ class ViT_SAM(DetectorBackbone):
             for param in module.parameters():
                 param.requires_grad_(False)
 
-    def set_causal_attention(self, is_causal: bool = True) -> None:
-        for b in self.body:
-            b.set_causal_attention(is_causal)
-
     def forward_features(self, x: torch.Tensor) -> torch.Tensor:
+        H, W = x.shape[-2:]
         x = self.patch_embed(x)
-        x = x + self.pos_embedding
+        x = x + self._get_pos_embed(H, W)
 
         x = self.body(x)
         x = self.neck(x.permute(0, 3, 1, 2))
@@ -410,9 +455,6 @@ class ViT_SAM(DetectorBackbone):
         x = self.forward_features(x)
         return self.features(x)
 
-    def set_dynamic_size(self, dynamic_size: bool = True) -> None:
-        assert dynamic_size is False, "Dynamic size not supported for this network"
-
     def adjust_size(self, new_size: tuple[int, int]) -> None:
         if new_size == self.size:
             return
@@ -530,6 +572,11 @@ class ViT_SAM(DetectorBackbone):
 
 
 # ViTDet (no neck)
+registry.register_model_config(
+    "vit_det_s16",
+    ViT_SAM,
+    config={"patch_size": 16, **SMALL, "window_size": 14, "global_attn_indexes": [2, 5, 8, 11]},
+)
 registry.register_model_config(
     "vit_det_m16_rms",
     ViT_SAM,
@@ -541,7 +588,6 @@ registry.register_model_config(
         "global_attn_indexes": [2, 5, 8, 11],
     },
 )
-
 registry.register_model_config(
     "vit_det_b16",
     ViT_SAM,

birder/scripts/train.py

@@ -581,10 +581,12 @@ def train(args: argparse.Namespace) -> None:
             if targets.ndim == 2:
                 targets = targets.argmax(dim=1)
 
-            train_accuracy.update(training_utils.accuracy(targets, outputs.detach()))
-            if train_topk is not None:
-                topk_val = training_utils.topk_accuracy(targets, outputs.detach(), topk=(top_k,))[0]
-                train_topk.update(topk_val)
+            if train_topk is None:
+                train_accuracy.update(training_utils.accuracy(targets, outputs.detach()))
+            else:
+                topk_values = training_utils.topk_accuracy(targets, outputs.detach(), topk=(1, top_k))
+                train_accuracy.update(topk_values[0])
+                train_topk.update(topk_values[1])
 
             # Write statistics
             if (i % args.log_interval == 0 and i > 0) or i == last_batch_idx:
@@ -682,10 +684,12 @@ def train(args: argparse.Namespace) -> None:
 
                 # Statistics
                 running_val_loss.update(val_loss.detach())
-                val_accuracy.update(training_utils.accuracy(targets, outputs), n=outputs.size(0))
-                if val_topk is not None:
-                    topk_val = training_utils.topk_accuracy(targets, outputs, topk=(top_k,))[0]
-                    val_topk.update(topk_val, n=outputs.size(0))
+                if val_topk is None:
+                    val_accuracy.update(training_utils.accuracy(targets, outputs), n=outputs.size(0))
+                else:
+                    topk_values = training_utils.topk_accuracy(targets, outputs, topk=(1, top_k))
+                    val_accuracy.update(topk_values[0], n=outputs.size(0))
+                    val_topk.update(topk_values[1], n=outputs.size(0))
 
                 # Update progress bar
                 progress.update(n=batch_size * args.world_size)

birder/scripts/train_capi.py

@@ -63,15 +63,21 @@ logger = logging.getLogger(__name__)
 
 
 class TrainCollator:
-    def __init__(self, mask_generator: Callable[[int], tuple[list[torch.Tensor], list[torch.Tensor]]]) -> None:
+    def __init__(
+        self, mask_generator: Callable[[int], tuple[list[torch.Tensor], list[torch.Tensor]]], n_predict: int
+    ) -> None:
         self.mask_generator = mask_generator
+        self.n_predict = n_predict
 
-    def __call__(self, batch: Any) -> tuple[torch.Tensor, torch.Tensor]:
+    def __call__(self, batch: Any) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
         B = len(batch)
         collated_batch = torch.utils.data.default_collate(batch)
         masks = self.mask_generator(B)
 
-        return (collated_batch, masks)
+        ids_keep = masking.get_ids_keep(masks)
+        predict_indices = masking.get_random_masked_indices(masks, self.n_predict)
+
+        return (collated_batch, ids_keep, predict_indices)
 
 
 # pylint: disable=too-many-locals,too-many-branches,too-many-statements
@@ -183,7 +189,7 @@ def train(args: argparse.Namespace) -> None:
     )
     n_masked = int(seq_len * args.mask_ratio)
     n_predict = int(n_masked * args.kept_mask_ratio)
-    mask_collator = TrainCollator(mask_generator)
+    mask_collator = TrainCollator(mask_generator, n_predict)
     training_transform = training_utils.get_training_transform(args)
     if args.use_fake_data is True:
         logger.warning("Using fake data")
@@ -491,17 +497,14 @@ def train(args: argparse.Namespace) -> None:
         else:
             batch_iter = enumerate(training_loader)
 
-        for i, ((_, images, _), masks) in batch_iter:
+        for i, ((_, images, _), ids_keep, predict_indices) in batch_iter:
             global_iter = ((epoch - 1) * epoch_num_batches) + i
             images = images.to(device, dtype=model_dtype, non_blocking=True)
-            masks = masks.to(device, dtype=model_dtype, non_blocking=True)
+            ids_keep = ids_keep.to(device, non_blocking=True)
+            predict_indices = predict_indices.to(device, non_blocking=True)
 
             optimizer_update = (i == last_batch_idx) or ((i + 1) % grad_accum_steps == 0)
 
-            # Mask handling
-            ids_keep = masking.get_ids_keep(masks)
-            predict_indices = masking.get_random_masked_indices(masks, n_predict)
-
             # Forward, backward and optimize
             with torch.amp.autocast("cuda", enabled=args.amp, dtype=amp_dtype):
                 selected_assignments, clustering_loss = teacher(images, None, predict_indices)

birder/scripts/train_detection.py

@@ -373,6 +373,7 @@ def train(args: argparse.Namespace) -> None:
         custom_keys_weight_decay=custom_keys_weight_decay,
         custom_layer_weight_decay=args.custom_layer_wd,
         layer_decay=args.layer_decay,
+        backbone_layer_decay=args.backbone_layer_decay,
         layer_decay_min_scale=args.layer_decay_min_scale,
         layer_decay_no_opt_scale=args.layer_decay_no_opt_scale,
         bias_lr=args.bias_lr,
@@ -948,7 +949,7 @@ def get_args_parser() -> argparse.ArgumentParser:
         help="treat all objects as a single class (binary detection: object vs background)",
     )
     training_cli.add_optimization_args(parser, default_batch_size=16)
-    training_cli.add_lr_wd_args(parser, backbone_lr=True)
+    training_cli.add_lr_wd_args(parser, backbone_lr=True, backbone_layer_decay=True)
     training_cli.add_lr_scheduler_args(parser)
     training_cli.add_training_schedule_args(parser)
     training_cli.add_ema_args(parser, default_ema_steps=1, default_ema_decay=0.9998)

birder/scripts/train_kd.py

@@ -708,10 +708,12 @@ def train(args: argparse.Namespace) -> None:
             if targets.ndim == 2:
                 targets = targets.argmax(dim=1)
 
-            train_accuracy.update(training_utils.accuracy(targets, outputs.detach()))
-            if train_topk is not None:
-                topk_val = training_utils.topk_accuracy(targets, outputs.detach(), topk=(top_k,))[0]
-                train_topk.update(topk_val)
+            if train_topk is None:
+                train_accuracy.update(training_utils.accuracy(targets, outputs.detach()))
+            else:
+                topk_values = training_utils.topk_accuracy(targets, outputs.detach(), topk=(1, top_k))
+                train_accuracy.update(topk_values[0])
+                train_topk.update(topk_values[1])
 
             # Write statistics
             if (i % args.log_interval == 0 and i > 0) or i == last_batch_idx:
@@ -804,10 +806,12 @@ def train(args: argparse.Namespace) -> None:
 
                 # Statistics
                 running_val_loss.update(val_loss.detach())
-                val_accuracy.update(training_utils.accuracy(targets, outputs), n=outputs.size(0))
-                if val_topk is not None:
-                    topk_val = training_utils.topk_accuracy(targets, outputs, topk=(top_k,))[0]
-                    val_topk.update(topk_val, n=outputs.size(0))
+                if val_topk is None:
+                    val_accuracy.update(training_utils.accuracy(targets, outputs), n=outputs.size(0))
+                else:
+                    topk_values = training_utils.topk_accuracy(targets, outputs, topk=(1, top_k))
+                    val_accuracy.update(topk_values[0], n=outputs.size(0))
+                    val_topk.update(topk_values[1], n=outputs.size(0))
 
                 # Update progress bar
                 progress.update(n=batch_size * args.world_size)

birder/version.py

@@ -1 +1 @@
-__version__ = "v0.4.1"
+__version__ = "v0.4.2"

{birder-0.4.1.dist-info → birder-0.4.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: birder
-Version: 0.4.1
+Version: 0.4.2
 Summary: An open-source computer vision framework for wildlife image analysis, featuring state-of-the-art models for species classification and detection.
 Author: Ofer Hasson
 License-Expression: Apache-2.0
@@ -26,7 +26,7 @@ License-File: LICENSE
 Requires-Dist: matplotlib>=3.9.0
 Requires-Dist: numpy>=2.2.0
 Requires-Dist: onnx>=1.18.0
-Requires-Dist: onnxscript~=0.5.7
+Requires-Dist: onnxscript~=0.6.0
 Requires-Dist: Pillow>=12.0.0
 Requires-Dist: polars>=1.31.0
 Requires-Dist: pyarrow>=20.0.0
@@ -48,7 +48,7 @@ Requires-Dist: black~=26.1.0; extra == "dev"
 Requires-Dist: build~=1.4.0; extra == "dev"
 Requires-Dist: bumpver~=2025.1131; extra == "dev"
 Requires-Dist: captum~=0.7.0; extra == "dev"
-Requires-Dist: coverage~=7.13.1; extra == "dev"
+Requires-Dist: coverage~=7.13.2; extra == "dev"
 Requires-Dist: debugpy; extra == "dev"
 Requires-Dist: flake8-pep585~=0.1.7; extra == "dev"
 Requires-Dist: flake8~=7.3.0; extra == "dev"

{birder-0.4.1.dist-info → birder-0.4.2.dist-info}/RECORD

@@ -1,6 +1,6 @@
 birder/__init__.py,sha256=qmAq9DNEsHDtvTTXJhp8Vzfrtc6ZR_TUjwjeyDW6VWc,869
 birder/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-birder/version.py,sha256=FyV6PQkiLDhwnTdAyvL5lOoA5FjP-pfJYkkmQC-1g6A,23
+birder/version.py,sha256=wir5g_QJ8Sj0VUgLyN6W--ATekZrCeDUAMMxqByjCgk,23
 birder/adversarial/__init__.py,sha256=PBdshTpEvO-aYZtD9kjkOF7Zor5t-P2V0wBwPhFr8_o,308
 birder/adversarial/base.py,sha256=CjUftBQjZTjDP-DW-Hgv5a81qOMPvf4V-pn0ZTsvwJU,3170
 birder/adversarial/deepfool.py,sha256=Om7C2OFk_uF7QcO4AO0pgjId-N0mIsU9HitpiWxk45o,6305
@@ -12,8 +12,8 @@ birder/common/cli.py,sha256=lOkAnp0L2YA8hAQk0orjluOHxEEK-CDJoRpeFKdyvWI,5427
 birder/common/fs_ops.py,sha256=Jshd7UdUSmQ7EJUaFm3ThlOKuSROm0qNC6Elt-DRc1Y,43410
 birder/common/lib.py,sha256=5w2YnQSvV-6b_k3j8U7XnY-_YzgTVEXkGGDg84mg6TQ,5907
 birder/common/masking.py,sha256=lsW1s5zm89A4LCQoVtq_7zc6Kg0A4j629wXnvpG12hs,14060
-birder/common/training_cli.py,sha256=rJEDAlu1q0OG1IyxdNvlN7PTtXy6NvXpzTcQjE1Qoxk,31850
-birder/common/training_utils.py,sha256=8TCG4fiwzLBzcMcGe7IhgezqSun0AaL9HswE1A6I3VY,50065
+birder/common/training_cli.py,sha256=REv_U3FUqgRVl9f5y1ghqTwCt3DGjyzfRGIVkCo3ZPk,32050
+birder/common/training_utils.py,sha256=3c3HP47WlQ9Oiq6rsgLrs6NqCfbgwSaFC69e6eqs-6U,53261
 birder/conf/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 birder/conf/settings.py,sha256=tfrE4mo_gGqtIfXm243tOZy69goYK5KA-VSBMhmw_WQ,1996
 birder/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -76,13 +76,13 @@ birder/model_registry/manifest.py,sha256=6gKayTBmnBm45fDTXTxZMmXnowS7VNaHq3fZSFq
 birder/model_registry/model_registry.py,sha256=myzRE4NMHqsVkokUO17FacJadHKKGxgjVIzFYAwqilk,9815
 birder/net/__init__.py,sha256=s2hWVFw_NcGHsvdsvqphtMkEzSGBgr1tSXPurAU0Zlw,6804
 birder/net/_rope_vit_configs.py,sha256=-mXwIwYTXAE0Umndjn1m4rdvaAGI9sq1Yh7tKF1kDn8,14018
-birder/net/_vit_configs.py,sha256=2dL7gj8pS9v2SXoEeR_1dlVPu4x5ZPQOUmoU6-718XA,14501
+birder/net/_vit_configs.py,sha256=6so9JKLDbec-kFZknNoNJ16pXqii2XkfflRykDgE0Ro,14663
 birder/net/alexnet.py,sha256=y1_KOnJyJWCQURin5DMu4wdJo3XGSzllBDwaariOW_c,2227
 birder/net/base.py,sha256=oUTtsg2T6P2gQq2JR8pp9AXImS1TlPCqIMmuSr8o66U,11234
 birder/net/biformer.py,sha256=ZGUGGGoZRSGAb7LwZ4VYyGPYJR1jkKGMnLEJVt29LUc,18978
-birder/net/cait.py,sha256=YABUTUROx3GkOMYBXeFgG83DgOgVw_WaoppLxarZqik,11289
+birder/net/cait.py,sha256=7K1y8R16Y61DWic8YygCA0Fv10G6EDEtnNik2LWKdJ8,11268
 birder/net/cas_vit.py,sha256=BWewoedWGU_Hdp5mSqX1Zu3e_fNvbzPeKw8C8llT4eU,12165
-birder/net/coat.py,sha256=iQleCPw2XVtQR9rYx83hW6aBDyO-xf1MTkwW5SNAGUw,25302
+birder/net/coat.py,sha256=allIZJ6ggYETYgzgbTuIXRB9VcI1tiIlgjjKHIE2PxY,25281
 birder/net/conv2former.py,sha256=2JBlsH6qy4Zj0wYyemSdXW1nBCd_nFQqiLLfOK1kJLo,10129
 birder/net/convmixer.py,sha256=2y3yDj5OGZGEBVF5Mgm9GuVab9BptHJX5M-8C0a-1nc,3515
 birder/net/convnext_v1.py,sha256=4b1FV3KoX2xV4kd_VW0AiYzh4kAqGcxkoNEwaZQZZ2Q,9247
@@ -94,12 +94,12 @@ birder/net/cspnet.py,sha256=0i3UO6f8PQlzBen0BJfdf1vhT2VjIvVrfRbZmUn6O7M,15465
 birder/net/cswin_transformer.py,sha256=nNf6C94Gxr_Hu0w1_W9paWkMfW1K2YfOCPtsxxSEasA,15152
 birder/net/darknet.py,sha256=oYbDi_HemdOzmwhLHKSQefEQ7fJmAdcoHqD2Rs1SAN8,4354
 birder/net/davit.py,sha256=qurj-baUX_oahKk0lQUhRIFU3EbpMGJEdaPIpivgwVg,16079
-birder/net/deit.py,sha256=9dN9rsQDxdCg59MtqhParXR_dmiR4ZBn5y-GSDxSb9I,10618
-birder/net/deit3.py,sha256=VwG4pHCskQoea5sTZMNVJ4pfx24mydnY8MylmdYA84w,18496
+birder/net/deit.py,sha256=bxl4A4J28ThuUrfBmOZothJiAh73Py3CtWVPplYIZcs,10631
+birder/net/deit3.py,sha256=WQk76_5dOrkz0wr-pC3a4Dy-x-2EYQ5E7YAOTUNqB48,18509
 birder/net/densenet.py,sha256=9hSloWLMZiaZtrithIEiSg_EfOtOLQPHbBctVSW6pBY,5972
 birder/net/dpn.py,sha256=OCIYOS30USJk-7L9geHD4kKQWS1stWXgh4atoX8KmDw,6685
 birder/net/edgenext.py,sha256=tEV3UkNOSSBmB_oE7x97EOWCvS_llwvPKG52p0Wkyr4,13799
-birder/net/edgevit.py,sha256=mduKFcvqT0WVJV6VHSxG2Rme5kTtPqJ6uH-RB5kDnQU,13138
+birder/net/edgevit.py,sha256=aaT5o6BL6bo5O8E_cCkMrqwSJ5eOQSaN1YGtiuz2QLo,13117
 birder/net/efficientformer_v1.py,sha256=Ya37V2DRIugYR8tVYPyxI0P2JWnHVmlZGExQiabMPd8,14291
 birder/net/efficientformer_v2.py,sha256=Z4qKYyFCE4IxIVOfEgyFSrP-4lKTV8EulAUeBdP_dUo,24881
 birder/net/efficientnet_lite.py,sha256=v6vTuG7p1Jit6fFfhieigxG_006V5zwpkDgf8oSpCF0,10991
@@ -107,10 +107,10 @@ birder/net/efficientnet_v1.py,sha256=xnmA0xJdE_1SCtHIeI3Y6Pb8qUmuVCmyWuN-bAg4LB0
 birder/net/efficientnet_v2.py,sha256=GTnWN5OVTT7HfNLAR2bSpeQ-tRb_GiMkyIFlkobOPYQ,13538
 birder/net/efficientvim.py,sha256=VlH6CHl3lz6JnwQidbLgzvsbIvwjmqT-VfQmVfFifXE,16858
 birder/net/efficientvit_mit.py,sha256=t4QyU9Q6hM29EClkaKni3U4vQDviIus4arySgTyQWeA,24345
-birder/net/efficientvit_msft.py,sha256=clajA5UJbAXWv9lEyHLNINOg0GiZNeJDis18gGVpuds,20144
+birder/net/efficientvit_msft.py,sha256=Nz_5D4LlHKZH70QUYFdPwsSASGcFVO1TPybtKMwqHu0,20138
 birder/net/fasternet.py,sha256=jEmAX8mvi939ocxk7XXKd5iatI1bgOee3vzR3YSfNvA,9019
 birder/net/fastvit.py,sha256=hr0m6b-IwrMN9BMbVmjxyR4kYiir64PCViV7YAM-PcU,38039
-birder/net/flexivit.py,sha256=fHwEoUr6BGMUtfkjGJxqpuyVwn7GLJYTTRi0Xa3diwE,24946
+birder/net/flexivit.py,sha256=LW-7iWpP1K8FaNju6vl7qNWhAeBhFqrnJ2O9_gSqolM,24959
 birder/net/focalnet.py,sha256=tIo8wvzv2LJhWzuRqXIkI6poRZZ3y-hJrXheT9zJzOo,20246
 birder/net/gc_vit.py,sha256=ea0sVCYH3w5yUIaWU31lrkiYD8xcK6CbjPkozXOPws4,23849
 birder/net/ghostnet_v1.py,sha256=P0dnO20evTd0-E7jvhb9RBRKOe5UgSygYfkmaYwoIP4,9048
@@ -119,7 +119,7 @@ birder/net/groupmixformer.py,sha256=DphJPX6nOavoQtMceNFAtlaA5iLd8v3qmRQDqSXOTOA,
 birder/net/hgnet_v1.py,sha256=fvQJdHMJnzWDJlD1XEuFSqSEkTZbZNJOFF65c893Yrg,14538
 birder/net/hgnet_v2.py,sha256=MUuJNEU2InsxeqFEp_xDljiH5ik27dx86mpsyZ-5gZc,10970
 birder/net/hiera.py,sha256=DhUFC6qRuUia2-Z7TrBMFt1jNcLM3IBlNGgUo9k5UeQ,28062
-birder/net/hieradet.py,sha256=2pFPsAhcNxCW-aZN4e7J9eCiv7NpgCUTenP0wTSYKts,20279
+birder/net/hieradet.py,sha256=tJHdWX-_F-EmnObWLLgLidw7D4iAYd1uOskqCRWt2_w,20281
 birder/net/hornet.py,sha256=E8F9Uljq71UtznHRTDrnYCBjhPVWjVHS6-Y3fT8Nfos,13907
 birder/net/iformer.py,sha256=CsWI23qnBl0CIIagwiDle-i-4_XWAh0uICIlGrJfs6I,17925
 birder/net/inception_next.py,sha256=PXVPeJszHZ7fJ2eOa3s1ChxpcOGSUjjnHi7qZ5WaZxE,10829
@@ -133,7 +133,7 @@ birder/net/lit_v1_tiny.py,sha256=FHndzM1lwMyWVjWINbb91ZB7hc_VgyaOlk2EUAFtZD8,118
 birder/net/lit_v2.py,sha256=E7PiTSdJHXfpbosHT37rh2fQUGwr_Uzru0c1FLOJS8M,13940
 birder/net/maxvit.py,sha256=w6BkzMfOESE9C9OOF2X1IQMbaODGHe5je9QQsju8S34,29762
 birder/net/metaformer.py,sha256=mia7kdlmyFRXhcTQjHKKdNoIzRWr1xh69N-mJPKk4Zg,30412
-birder/net/mnasnet.py,sha256=uGjkWmr57-QeKhd33S4Ko_fCMXMC2LgGg6aqTNHOQ4M,9902
+birder/net/mnasnet.py,sha256=PH7YpOy1Va93oPickYwv4PN3NRr54gRLy-wfTTSbCGw,9889
 birder/net/mobilenet_v1.py,sha256=gbqBkKvCadZAngvTP3psh1bN6zjmP3aQFfCkG3oB5qU,5325
 birder/net/mobilenet_v2.py,sha256=cDTRkALF07OSc3xYUKqNsfrHnUqJTor0yQ_Hqtf3LJg,7698
 birder/net/mobilenet_v3.py,sha256=YyWsbmMjGQ9Nwp5NkMdPB3ns3BwRTEdbHuQT3kkt5rw,13678
@@ -160,14 +160,14 @@ birder/net/resmlp.py,sha256=NVrTRCP4q_mWhbJOywX5f_lqNhqKvhwiloUEoBZZDGc,4687
 birder/net/resnest.py,sha256=_ql3dcLkqkhT91E-9-M0SiGty9uknwfns3uy4KZ3GkE,10104
 birder/net/resnet_v1.py,sha256=TIv8TjywneOFsYxJQhLrwCgnOTS1V-y_-pu2qjU4tu8,14546
 birder/net/resnet_v2.py,sha256=0QnKggNzpIwhAdI4gLej04VjuJGNVwcW9rPrIdtbnfM,9286
-birder/net/resnext.py,sha256=mdDfUe0YxiDiRn8qhvR324DaMjFmefoT-LO4Rc7HZ6k,10265
-birder/net/rope_deit3.py,sha256=asj7W0BvTkiHC1GIk3bb-suh_whA3hwUa7WVTQzssrw,22386
-birder/net/rope_flexivit.py,sha256=mTtIL6X0yJwX9iR-Fgf9rgsHoR0vB_4hAzflyoA9OHI,27413
-birder/net/rope_vit.py,sha256=iRQ_TVSe8qCXTlnUJx0dFyQ9Z6A3CneHRBnlTU9R6n4,43280
+birder/net/resnext.py,sha256=ENa3pWL2mh_zYgk3OIYCBQiFltQLim27MfFRI2BBBNs,10252
+birder/net/rope_deit3.py,sha256=txncunQ9XdKyfwgfDsNs0AigeF4P4y66xEHXuOkrIUo,22399
+birder/net/rope_flexivit.py,sha256=YpBOjuvFImPU5uCna-Cl1KpWMtXs-xMBKsjJ93wdU_4,27426
+birder/net/rope_vit.py,sha256=PCgmwIxJrrq5ctIwKF9MRtADKxSIb2ZncKi0EQY21zA,43293
 birder/net/sequencer2d.py,sha256=WcPlqawg6oQ9gg6h-WJmM25uxnDXA1wsfvJP53UUj60,7416
 birder/net/shufflenet_v1.py,sha256=BAJH4fb8R9CeqHqAcczPF94H9ipjsyPPl2Da_cbzir0,8129
 birder/net/shufflenet_v2.py,sha256=HNF33OmFlLgPN-VilZ9-JF78NJ60MiV2GlSwV8myv7c,6822
-birder/net/simple_vit.py,sha256=Uh9SiRnM7MgJM5oJbPB8yP9KGmgtOiLqVVV3mDq9GxI,11006
+birder/net/simple_vit.py,sha256=PadSMtNnMJdhB0hHaXSrHkXTR92sBvGrg0tFMQko1Rg,11019
 birder/net/smt.py,sha256=wIY-RtKP6W_3WlAl-C0N_boybRJ4ZMxTSCXWDnS40-c,16198
 birder/net/squeezenet.py,sha256=bUhtqQd9V43vKEcUJG03JkxxCX7iBS_QHFjz497MQP4,3505
 birder/net/squeezenext.py,sha256=2nDG9E02UNlaquKgT-U6JgdyaRd_Hlwbesk81iIfbWc,6026
@@ -181,28 +181,29 @@ birder/net/uniformer.py,sha256=FYDzOAu4IvRiTVR6DQ1anooRQKiQTxyKFIIrzKAhsMc,12674
 birder/net/van.py,sha256=IAymYsIEK3zzST4KS6iU689b29Gm8TCwLFV2BjPyM0g,10671
 birder/net/vgg.py,sha256=-iTVoBXat_6jo8gnpA3Jk4yEBdMKLVbY4iMe_KJ-0Fw,3226
 birder/net/vgg_reduced.py,sha256=1NmjS4KSMvSEU3U090gf82bvWkyYDssEglaXNauCw4A,3356
-birder/net/vit.py,sha256=jeuxRrN4iT5Ue0ukO4Sx5mxrLGCnC9-k4WBuyS11Mnk,39229
-birder/net/vit_parallel.py,sha256=N-QGz7v_MeuX8FNjpnyl_ha9qD5F3X5G94lPdwPgWRc,23672
-birder/net/vit_sam.py,sha256=AckImaaux5cbw08ZvtEY1XJa9UEHCxSPA4ju3dUilXU,21157
+birder/net/vit.py,sha256=8C9wKcin0llrZTtArz2IiQBd5TqM2aj4No_u3fBel64,39971
+birder/net/vit_parallel.py,sha256=GzGCs0_OFjZEuViJMqK6LBRNX6-PWMjWh8vwdNjkIr8,23685
+birder/net/vit_sam.py,sha256=NMsUaiSyfahGQvMiMb-nFKwLlCHSr8gGPoKaHarNTMM,22930
 birder/net/vovnet_v1.py,sha256=4Wgm1mALHpTJlhcrczKPVPkzDh3RNTGTty1FLzRK1EM,7685
 birder/net/vovnet_v2.py,sha256=8kuLm3bURn_I3j0m9YmiJI4TohuWyB-UBlefWIa694g,9714
 birder/net/wide_resnet.py,sha256=d6yn-Fb67zCGLMhQLV09gik5Otah6nCDronm4betcPA,5972
 birder/net/xception.py,sha256=58rncfGZ2U3-aGZjBABhPenRBZl8cvttrtSc0iGMVck,6913
 birder/net/xcit.py,sha256=0WH7Zex1dVpUbAZ7aiwyhhTNDiJd33KtFZGzno3cEQk,19747
-birder/net/detection/__init__.py,sha256=2kusDH9kBe645DNwUQE59ArQZxphhYeIrhWWyLPTPUU,1105
+birder/net/detection/__init__.py,sha256=gkPYjk8x2L095DHRJPogYNCrzhIzoZaNwh6_AMPWZFE,1169
 birder/net/detection/_yolo_anchors.py,sha256=mbSWosbgfvw6KJ4AcfXV7LoRgTuYgX3yg4VH2zajLD0,6911
 birder/net/detection/base.py,sha256=tHyXyGggKPjUaDiKnftYe6fNjZBK2TokenvgMVYhLf8,26248
-birder/net/detection/deformable_detr.py,sha256=mcBT2h-SzA1n7nq0X4i10PGD08lYmL_rqagK4LLF61E,34646
-birder/net/detection/detr.py,sha256=Fa8EVTH-lWrJgW-Y0Dc6UOmuPqyJqOsg0uMbO_Z_t6M,20670
+birder/net/detection/deformable_detr.py,sha256=lE8VZiONTYaDKlc-OU2PQbs9w49XUHJ4C1NnfBwkyMM,34616
+birder/net/detection/detr.py,sha256=76AlLiljL9oFkNrJ3ZblxOkD5tAC7oGzwGFmqhuThYE,20680
 birder/net/detection/efficientdet.py,sha256=1IrqoWbL6AX3g0jcdKqe010VHeD6GWEmEp9qmLxnNSE,29038
 birder/net/detection/faster_rcnn.py,sha256=JOI4nlg5ss5_nvhklBhg7EBZu7e40zxmd-9i2avVCqg,34338
 birder/net/detection/fcos.py,sha256=Qhe5y5ito4UZ9TtquWaVyslKWmv4xy56dKGRECkRf18,21867
-birder/net/detection/plain_detr.py,sha256=SQ_0vWcegHNaB0PDsO4cZq7LLcMgC4wRHE0v_DgEzpc,32857
-birder/net/detection/retinanet.py,sha256=PjkEpoKiAN6IkLleHaIFFQyOQbOFMXZEpn9lM-C8eeM,19076
-birder/net/detection/rt_detr_v1.py,sha256=C0cONbzEtToVm0dQl2IeKBkWrL0CTxCQr7EELdsGiaA,46544
-birder/net/detection/rt_detr_v2.py,sha256=3bU25aQgD-XcsTdgQx0TE6RfpU7N9RfZMdiSkk6oqB8,46355
+birder/net/detection/lw_detr.py,sha256=sA0Uj6B-CYokt9yhufRf3Ks24yUWtcGi7L4IbruY2fk,50516
+birder/net/detection/plain_detr.py,sha256=8DgpyYXX8P_s4gwPBhzkTYXwZZ5IBlRiDNqYcU-8Dd0,32916
+birder/net/detection/retinanet.py,sha256=uTUPOmvHOJ9R_I4NRqUr92AMZIeb-3TP_3ZkNFNyHqs,19070
+birder/net/detection/rt_detr_v1.py,sha256=-LU3ZdQuRled6Q2LWTk4hy4kKyu8sTMnHGk9TBkID48,46444
+birder/net/detection/rt_detr_v2.py,sha256=f0Mqvo_se0dXypfm-3OYQYoPRPDHm4o8hJvE-tCSwHA,45502
 birder/net/detection/ssd.py,sha256=SxTU151Fn8DmylFkqVp3_6F29BvZkQHejDDti3A554A,20339
-birder/net/detection/ssdlite.py,sha256=pv2ychUWUu-UWxVs6_RXMdjEiIPfkBoscschCGNDXXI,7241
+birder/net/detection/ssdlite.py,sha256=0LwDiiS20xNjkiLvEyyW-cBm32py6fdsIBqPDxkvsKc,7225
 birder/net/detection/vitdet.py,sha256=bpyX0W_q0tBqxQ7hptHnn8Fm5y0DkXvRQQBiRcm-9RE,4424
 birder/net/detection/yolo_v2.py,sha256=WpA5ZOtbzAFsUqPQDplXAJxMmDpVVGnBO7KTvzaEgrs,19949
 birder/net/detection/yolo_v3.py,sha256=INTX3ciZ678PighKN6sNp8Xs3iWZT4aXYcP4ANjS8mo,27777
@@ -250,20 +251,20 @@ birder/scripts/benchmark.py,sha256=BNxrZDSs9bf7xH0AManKPJk9RTGmCqiUP95jBYx8wUw,1
 birder/scripts/evaluate.py,sha256=CbCWK2TR19bEdH0wPO3iCbS0EjZWxfsBAJ42hDD6fKI,6791
 birder/scripts/predict.py,sha256=ryaWcy8xVN16MnLxf2fgxSMa6Wyb7CHi76HgVMCGoA4,30314
 birder/scripts/predict_detection.py,sha256=n3-EJ3Lyg1cLOAgEEj1uyfqHGJRJlkMKKU28WY3i7Ts,17563
-birder/scripts/train.py,sha256=gvR89QDW78QzFUOY-yhCWVp4sOW4NHAKB3bT8f3wOrs,38140
+birder/scripts/train.py,sha256=u46D7MDApmTg9GR0g4ANAbMWiEbO6MDwvzq5WwtZysY,38325
 birder/scripts/train_barlow_twins.py,sha256=pUIAoS4ZbVnL7nuA-uQFHfxwnuJrZtM1xe0RC-C8uRs,23240
 birder/scripts/train_byol.py,sha256=cD3ItGOSmF2-EvVzt7qHNO95wk0YVWusN05wnXPs18k,24207
-birder/scripts/train_capi.py,sha256=Af7CyyUVRDWnroEAPN1zpPq9ZWv_qrXZ8911iJBmU70,32124
+birder/scripts/train_capi.py,sha256=agED-24MMOCsgtlebG8KEl05l17WFDadLFCPLbgWCQA,32286
 birder/scripts/train_data2vec.py,sha256=6UC3TDcnelz2R2sJcdiBsRv_rLtAs0rs79Am1EqgLuM,24651
 birder/scripts/train_data2vec2.py,sha256=OnagbpMyr4YHW6Cuo5I7oW2bY-877MlUs0w5PxZ9z_k,25812
-birder/scripts/train_detection.py,sha256=dEAxfS8ewQOK_j6gt-GazA88mOePTNUV2qBvd3A_ae8,41556
+birder/scripts/train_detection.py,sha256=Jv4oUH-zfEpLtH7ysnLuK3_i8p1rdlVpxALI6vDi9Cw,41639
 birder/scripts/train_dino_v1.py,sha256=atZjI_51RX0rWWO7ptKfduti3CF5F-6SNpPRhVdKOzM,32189
 birder/scripts/train_dino_v2.py,sha256=7V3SuyMpcQvLyeBQmriNFCguG2w3aYSRQq9z7pBr6eM,48251
 birder/scripts/train_dino_v2_dist.py,sha256=eOr52W4rROpQScPrg03LhTuAAXRrCCaHzIvHmCsPoD8,49635
 birder/scripts/train_franca.py,sha256=-OcEjPzOYWEVeFfGr4Dl1_Cee7hFNeBWzKctrHJ9VJc,43736
 birder/scripts/train_i_jepa.py,sha256=_C3N32gFFTF512eQgGsZ8tqSsz1vsCauot5bPq9aqrE,28127
 birder/scripts/train_ibot.py,sha256=Pw-LZs2_-PWJj9iyidgdNdbCGj3QoIrQmlLDLctlkS8,33635
-birder/scripts/train_kd.py,sha256=DmNZkWIdB1OuZqUYqII0XZX-ZDJn4I14A-LMln-bL9U,44476
+birder/scripts/train_kd.py,sha256=1_ynCs0vDln57Om3YLw4txlExrYOLWQ7I5uojSPVyqA,44661
 birder/scripts/train_mim.py,sha256=Hu63BxHjGoOUgPPNElSuCdNT2-QYOoE0wWrR3E3fPxs,24251
 birder/scripts/train_mmcr.py,sha256=laXYFvHTNwB1r--4elKG__7_IOkIQQkrZYzY6PrZsnA,24230
 birder/scripts/train_rotnet.py,sha256=Td7WDkPmooP2TaedcDp1mIvQzRQHQmwA0TQN1wh2Zy4,24474
@@ -292,9 +293,9 @@ birder/tools/stats.py,sha256=TapTDidm4Vhtb_q9Wm_v7V9VHchC0qaOmtPtLuoLaAk,8375
 birder/tools/verify_coco.py,sha256=lMCc9X-P-odMEN6GkNpV_TDY7grze11RkPMs-Z_Uy-4,3717
 birder/tools/verify_directory.py,sha256=KbA97CX2HrwdL97HoKmSO8o3vFdP19TT7zHkulMp8L4,3284
 birder/tools/voc_to_coco.py,sha256=ktCviq1Bu9un27mWWPzsr5RolrOpwbFNID5EubY2C1k,5249
-birder-0.4.1.dist-info/licenses/LICENSE,sha256=tQ5GIY23TCjwH64fidwWBp8JdK2QbPVl1UePNd-5m5U,11341
-birder-0.4.1.dist-info/METADATA,sha256=gdJ2ADaW_-Qd8Pn_D1pvWHkXMcc41pYueN2FJpvvEjM,12696
-birder-0.4.1.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
-birder-0.4.1.dist-info/entry_points.txt,sha256=nikWKO1k2GISDo1hsv2T-37NCwD4dtb3PXRwPkOZ86w,128
-birder-0.4.1.dist-info/top_level.txt,sha256=ZRzmYwYDWwwOKFYgW4k0vUysaazrAJDLnpIjVH6cwtY,7
-birder-0.4.1.dist-info/RECORD,,
+birder-0.4.2.dist-info/licenses/LICENSE,sha256=tQ5GIY23TCjwH64fidwWBp8JdK2QbPVl1UePNd-5m5U,11341
+birder-0.4.2.dist-info/METADATA,sha256=IrahFdkSe2V4zwqbT6ZmKNTuG18vBbtlTKuoDJqbvp4,12696
+birder-0.4.2.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+birder-0.4.2.dist-info/entry_points.txt,sha256=nikWKO1k2GISDo1hsv2T-37NCwD4dtb3PXRwPkOZ86w,128
+birder-0.4.2.dist-info/top_level.txt,sha256=ZRzmYwYDWwwOKFYgW4k0vUysaazrAJDLnpIjVH6cwtY,7
+birder-0.4.2.dist-info/RECORD,,

{birder-0.4.1.dist-info → birder-0.4.2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.10.1)
+Generator: setuptools (80.10.2)
 Root-Is-Purelib: true
 Tag: py3-none-any