Upload model

README.md

@@ -1,13 +1,12 @@
 ---
 library_name: transformers
 tags: []
-
 co2_eq_emissions:
   emissions: 240
-  source: "https://calculator.green-algorithms.org/"
-  training_type: "pre-training"
-  geographical_location: "Switzerland"
-  hardware_used: "1 v100 GPU"
+  source: https://calculator.green-algorithms.org/
+  training_type: pre-training
+  geographical_location: Switzerland
+  hardware_used: 1 v100 GPU
 ---
 
 # Model Card for Model ID

config.json

@@ -1,85 +1,85 @@
-{
-  "architectures": [
-    "CetaceanClassifierModelForImageClassification"
-  ],
-  "aug": {
-    "crop_l": 0.75,
-    "crop_r": 1.3333333333333333,
-    "crop_scale": 0.9,
-    "p_affine": 0.5,
-    "p_blur": 0.05,
-    "p_bright_contrast": 0.5,
-    "p_cutout": 0.05,
-    "p_downscale": 0.0,
-    "p_gray": 0.1,
-    "p_noise": 0.05,
-    "p_posterize": 0.2,
-    "p_rain": 0.05,
-    "p_shuffle": 0.3,
-    "p_snow": 0.1,
-    "rotate": 15,
-    "shear": 3,
-    "translate": 0.25
-  },
-  "auto_map": {
-    "AutoConfig": "configuration_cetacean_classifier.CetaceanClassifierConfig",
-    "AutoModelForImageClassification": "modeling_cetacean_classifier.CetaceanClassifierModelForImageClassification"
-  },
-  "batch_size": 8,
-  "bbox_conf_threshold": 0.01,
-  "bboxes": {
-    "backfin": 0.15,
-    "detic": 0.05,
-    "fullbody": 0.6,
-    "fullbody_charm": 0.15,
-    "none": 0.05
-  },
-  "global_pool": {
-    "arch": "GeM",
-    "p": 3,
-    "train": false
-  },
-  "image_size": [
-    768,
-    768
-  ],
-  "loss_fn": "CrossEntropy",
-  "loss_id_ratio": 0.437338,
-  "lr_backbone": 0.0016,
-  "lr_decay_scale": 0.01,
-  "lr_head": 0.016,
-  "margin_coef_id": 0.27126,
-  "margin_coef_species": 0.226253,
-  "margin_cons_id": 0.05,
-  "margin_cons_species": 0.05,
-  "margin_power_id": -0.364399,
-  "margin_power_species": -0.720133,
-  "max_epochs": 30,
-  "model_name": "tf_efficientnetv2_m",
-  "model_type": "cetaceanet",
-  "n_center_id": 2,
-  "n_center_species": 2,
-  "n_data": -1,
-  "n_splits": -1,
-  "normalization": "batchnorm",
-  "num_classes": 15587,
-  "num_species_classes": 26,
-  "optimizer": "AdamW",
-  "out_indices": [
-    3,
-    4
-  ],
-  "pretrained": true,
-  "pseudo_conf_threshold": 0.6,
-  "pseudo_label": "pseudo_labels/round2.csv",
-  "s_id": 20.9588,
-  "s_species": 33.1383,
-  "test_bboxes": [
-    "fullbody",
-    "fullbody_charm"
-  ],
-  "torch_dtype": "float32",
-  "transformers_version": "4.46.0",
-  "val_bbox": "fullbody",
-  "warmup_steps_ratio": 0.2
-}
+{
+  "architectures": [
+    "CetaceanClassifierModelForImageClassification"
+  ],
+  "aug": {
+    "crop_l": 0.75,
+    "crop_r": 1.3333333333333333,
+    "crop_scale": 0.9,
+    "p_affine": 0.5,
+    "p_blur": 0.05,
+    "p_bright_contrast": 0.5,
+    "p_cutout": 0.05,
+    "p_downscale": 0.0,
+    "p_gray": 0.1,
+    "p_noise": 0.05,
+    "p_posterize": 0.2,
+    "p_rain": 0.05,
+    "p_shuffle": 0.3,
+    "p_snow": 0.1,
+    "rotate": 15,
+    "shear": 3,
+    "translate": 0.25
+  },
+  "auto_map": {
+    "AutoConfig": "configuration_cetacean_classifier.CetaceanClassifierConfig",
+    "AutoModelForImageClassification": "modeling_cetacean_classifier.CetaceanClassifierModelForImageClassification"
+  },
+  "batch_size": 8,
+  "bbox_conf_threshold": 0.01,
+  "bboxes": {
+    "backfin": 0.15,
+    "detic": 0.05,
+    "fullbody": 0.6,
+    "fullbody_charm": 0.15,
+    "none": 0.05
+  },
+  "global_pool": {
+    "arch": "GeM",
+    "p": 3,
+    "train": false
+  },
+  "image_size": [
+    768,
+    768
+  ],
+  "loss_fn": "CrossEntropy",
+  "loss_id_ratio": 0.437338,
+  "lr_backbone": 0.0016,
+  "lr_decay_scale": 0.01,
+  "lr_head": 0.016,
+  "margin_coef_id": 0.27126,
+  "margin_coef_species": 0.226253,
+  "margin_cons_id": 0.05,
+  "margin_cons_species": 0.05,
+  "margin_power_id": -0.364399,
+  "margin_power_species": -0.720133,
+  "max_epochs": 30,
+  "model_name": "tf_efficientnetv2_m",
+  "model_type": "cetaceanet",
+  "n_center_id": 2,
+  "n_center_species": 2,
+  "n_data": -1,
+  "n_splits": -1,
+  "normalization": "batchnorm",
+  "num_classes": 15587,
+  "num_species_classes": 26,
+  "optimizer": "AdamW",
+  "out_indices": [
+    3,
+    4
+  ],
+  "pretrained": true,
+  "pseudo_conf_threshold": 0.6,
+  "pseudo_label": "pseudo_labels/round2.csv",
+  "s_id": 20.9588,
+  "s_species": 33.1383,
+  "test_bboxes": [
+    "fullbody",
+    "fullbody_charm"
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.46.0",
+  "val_bbox": "fullbody",
+  "warmup_steps_ratio": 0.2
+}

configuration_cetacean_classifier.py

@@ -1,12 +1,12 @@
-from transformers import PretrainedConfig
-from typing import List
-
-
-class CetaceanClassifierConfig(PretrainedConfig):
-    model_type = "cetaceanet"
-
-    def __init__(
-        self,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
+from transformers import PretrainedConfig
+from typing import List
+
+
+class CetaceanClassifierConfig(PretrainedConfig):
+    model_type = "cetaceanet"
+
+    def __init__(
+        self,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c9afc61a269bf406f5b23389c57e4efe365eb4b67aa62730b731916fb62b6f0
+oid sha256:faf081f38c0224790f5819d763475e9827667fa50af1a86a61e233c52876ae11
 size 296028464

modeling_cetacean_classifier.py

@@ -1,71 +1,91 @@
-from transformers import PreTrainedModel
-from PIL import Image
-import numpy as np
-import torch
-
-from .configuration_cetacean_classifier import CetaceanClassifierConfig
-from .train import SphereClassifier
-
-
-WHALE_CLASSES = np.array(
-    [
-        "beluga",
-        "blue_whale",
-        "bottlenose_dolphin",
-        "brydes_whale",
-        "commersons_dolphin",
-        "common_dolphin",
-        "cuviers_beaked_whale",
-        "dusky_dolphin",
-        "false_killer_whale",
-        "fin_whale",
-        "frasiers_dolphin",
-        "gray_whale",
-        "humpback_whale",
-        "killer_whale",
-        "long_finned_pilot_whale",
-        "melon_headed_whale",
-        "minke_whale",
-        "pantropic_spotted_dolphin",
-        "pygmy_killer_whale",
-        "rough_toothed_dolphin",
-        "sei_whale",
-        "short_finned_pilot_whale",
-        "southern_right_whale",
-        "spinner_dolphin",
-        "spotted_dolphin",
-        "white_sided_dolphin",
-    ]
-)
-
-
-class CetaceanClassifierModelForImageClassification(PreTrainedModel):
-    config_class = CetaceanClassifierConfig
-
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.model = SphereClassifier(cfg=config.to_dict())
-
-        # load_from_checkpoint("cetacean_classifier/last.ckpt")
-        # self.model = SphereClassifier.load_from_checkpoint("cetacean_classifier/last.ckpt")
-
-        self.model.eval()
-
-    def preprocess_image(self, img: Image) -> torch.Tensor:
-        image_resized = img.resize((480, 480))
-        image_resized = np.array(image_resized)[None]
-        image_resized = np.transpose(image_resized, [0, 3, 2, 1])
-        image_tensor = torch.Tensor(image_resized)
-        return image_tensor
-
-    def forward(self, img: Image, labels=None):
-        tensor = self.preprocess_image(img)
-        head_id_logits, head_species_logits = self.model(tensor)
-        head_species_logits = head_species_logits.detach().numpy()
-        sorted_idx = head_species_logits.argsort()[0]
-        sorted_idx = np.array(list(reversed(sorted_idx)))
-        top_three_logits = sorted_idx[:3]
-        top_three_whale_preds = WHALE_CLASSES[top_three_logits]
-
-        return {"predictions": top_three_whale_preds}
+import albumentations as A
+from transformers import PreTrainedModel
+from PIL import Image
+import numpy as np
+import torch
+import cv2
+from configuration_cetacean_classifier import CetaceanClassifierConfig
+from train import SphereClassifier
+
+
+WHALE_CLASSES = np.array(
+    [
+        "beluga",
+        "blue_whale",
+        "bottlenose_dolphin",
+        "brydes_whale",
+        "commersons_dolphin",
+        "common_dolphin",
+        "cuviers_beaked_whale",
+        "dusky_dolphin",
+        "false_killer_whale",
+        "fin_whale",
+        "frasiers_dolphin",
+        "gray_whale",
+        "humpback_whale",
+        "killer_whale",
+        "long_finned_pilot_whale",
+        "melon_headed_whale",
+        "minke_whale",
+        "pantropic_spotted_dolphin",
+        "pygmy_killer_whale",
+        "rough_toothed_dolphin",
+        "sei_whale",
+        "short_finned_pilot_whale",
+        "southern_right_whale",
+        "spinner_dolphin",
+        "spotted_dolphin",
+        "white_sided_dolphin",
+    ]
+)
+
+
+class CetaceanClassifierModelForImageClassification(PreTrainedModel):
+    config_class = CetaceanClassifierConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.model = SphereClassifier(cfg=config.to_dict())
+
+        # load_from_checkpoint("cetacean_classifier/last.ckpt")
+        # self.model = SphereClassifier.load_from_checkpoint("cetacean_classifier/last.ckpt")
+
+        self.model.eval()
+        self.config = config
+        self.transforms = self.make_transforms(data_aug=True)
+    
+    def make_transforms(self, data_aug: bool):
+        augments = []
+        if data_aug:
+            aug = self.config.aug
+            augments = [
+                A.RandomResizedCrop(
+                    self.config.image_size[0],
+                    self.config.image_size[1],
+                    scale=(aug["crop_scale"], 1.0),
+                    ratio=(aug["crop_l"], aug["crop_r"]),
+                ),]
+        return A.Compose(augments)
+
+    def preprocess_image(self, img) -> torch.Tensor:        
+        rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        image = cv2.resize(rgb, self.config.image_size, interpolation=cv2.INTER_CUBIC)
+        image =  self.transforms(image=image)["image"]
+        return torch.Tensor(image).transpose(2, 0).unsqueeze(0)
+        #image_resized = img.resize((480, 480))
+        #image_resized = np.array(image_resized)[None]
+        #image_resized = np.transpose(image_resized, [0, 3, 2, 1])
+        #image_tensor = torch.Tensor(image_resized)
+        #return image_tensor
+
+    def forward(self, img, labels=None):
+        tensor = self.preprocess_image(img)
+        head_id_logits, head_species_logits = self.model(tensor)
+        head_species_logits = head_species_logits.detach().numpy()
+        sorted_idx = head_species_logits.argsort()[0]
+        sorted_idx = np.array(list(reversed(sorted_idx)))
+        top_three_logits = sorted_idx[:3]
+        top_three_whale_preds = WHALE_CLASSES[top_three_logits]
+
+        return {"predictions": top_three_whale_preds}