tiny rotated object detection with weights

Author: bsridatta

bsridatta.egg-info/PKG-INFO

@@ -0,0 +1,11 @@
+Metadata-Version: 1.2
+Name: bsridatta
+Version: 0.0.1
+Summary: UNKNOWN
+Home-page: UNKNOWN
+Author: Sri Datta Budaraju
+Author-email: b.sridatta@gmail.com
+License: UNKNOWN
+Description: UNKNOWN
+Platform: UNKNOWN
+Requires-Python: >=3.7

bsridatta.egg-info/SOURCES.txt

@@ -0,0 +1,9 @@
+setup.py
+bsridatta.egg-info/PKG-INFO
+bsridatta.egg-info/SOURCES.txt
+bsridatta.egg-info/dependency_links.txt
+bsridatta.egg-info/top_level.txt
+src/__init__.py
+src/dataloader.py
+src/dataset.py
+src/train.py

bsridatta.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

bsridatta.egg-info/top_level.txt

@@ -0,0 +1 @@
+src

model.hdf5

@@ -0,0 +1,67 @@
+
+----------------------------------------------------------------
+        Layer (type)               Output Shape         Param #
+================================================================
+            Conv2d-1          [-1, 8, 200, 200]              72
+       BatchNorm2d-2          [-1, 8, 200, 200]              16
+         MaxPool2d-3          [-1, 8, 100, 100]               0
+            Conv2d-4         [-1, 16, 100, 100]           1,152
+       BatchNorm2d-5         [-1, 16, 100, 100]              32
+         MaxPool2d-6           [-1, 16, 50, 50]               0
+            Conv2d-7           [-1, 32, 25, 25]           4,608
+       BatchNorm2d-8           [-1, 32, 25, 25]              64
+              Mish-9           [-1, 32, 25, 25]               0
+           Conv2d-10           [-1, 32, 25, 25]           9,216
+      BatchNorm2d-11           [-1, 32, 25, 25]              64
+           Conv2d-12           [-1, 32, 25, 25]             512
+      BatchNorm2d-13           [-1, 32, 25, 25]              64
+             Mish-14           [-1, 32, 25, 25]               0
+        ConvBlock-15           [-1, 32, 25, 25]               0
+           Conv2d-16           [-1, 64, 13, 13]          18,432
+      BatchNorm2d-17           [-1, 64, 13, 13]             128
+             Mish-18           [-1, 64, 13, 13]               0
+           Conv2d-19           [-1, 64, 13, 13]          36,864
+      BatchNorm2d-20           [-1, 64, 13, 13]             128
+           Conv2d-21           [-1, 64, 13, 13]           2,048
+      BatchNorm2d-22           [-1, 64, 13, 13]             128
+             Mish-23           [-1, 64, 13, 13]               0
+        ConvBlock-24           [-1, 64, 13, 13]               0
+           Conv2d-25            [-1, 128, 7, 7]          73,728
+      BatchNorm2d-26            [-1, 128, 7, 7]             256
+             Mish-27            [-1, 128, 7, 7]               0
+           Conv2d-28            [-1, 128, 7, 7]         147,456
+      BatchNorm2d-29            [-1, 128, 7, 7]             256
+           Conv2d-30            [-1, 128, 7, 7]           8,192
+      BatchNorm2d-31            [-1, 128, 7, 7]             256
+             Mish-32            [-1, 128, 7, 7]               0
+        ConvBlock-33            [-1, 128, 7, 7]               0
+           Conv2d-34            [-1, 256, 4, 4]         294,912
+      BatchNorm2d-35            [-1, 256, 4, 4]             512
+             Mish-36            [-1, 256, 4, 4]               0
+           Conv2d-37            [-1, 256, 4, 4]         589,824
+      BatchNorm2d-38            [-1, 256, 4, 4]             512
+           Conv2d-39            [-1, 256, 4, 4]          32,768
+      BatchNorm2d-40            [-1, 256, 4, 4]             512
+             Mish-41            [-1, 256, 4, 4]               0
+        ConvBlock-42            [-1, 256, 4, 4]               0
+           Conv2d-43            [-1, 256, 4, 4]          65,792
+           Conv2d-44            [-1, 256, 7, 7]          33,024
+           Conv2d-45          [-1, 256, 13, 13]          16,640
+           Conv2d-46          [-1, 256, 25, 25]           8,448
+           Conv2d-47          [-1, 256, 25, 25]         590,080
+AdaptiveAvgPool2d-48            [-1, 256, 1, 1]               0
+AdaptiveAvgPool2d-49            [-1, 256, 1, 1]               0
+          Flatten-50                  [-1, 256]               0
+           Linear-51                    [-1, 1]             257
+          Flatten-52                  [-1, 256]               0
+           Linear-53                    [-1, 5]           1,285
+================================================================
+Total params: 1,938,238
+Trainable params: 1,938,238
+Non-trainable params: 0
+----------------------------------------------------------------
+Input size (MB): 0.15
+Forward/backward pass size (MB): 13.97
+Params size (MB): 7.39
+Estimated Total Size (MB): 21.52
+----------------------------------------------------------------

reports/FPN-summary ordered table.png

@@ -0,0 +1,11 @@
+matplotlib==3.3.1
+numpy==1.18.5
+tqdm==4.48.2
+Shapely==1.7.0
+scikit_image==0.16.2
+torch==1.6.0
+torchvision==0.7.0
+tensorflow==2.3.0
+Pillow==7.2.0
+pytest==6.0.2
+torchsummary==1.5.1

reports/FPN_torchsummary.txt

@@ -0,0 +1,10 @@
+import setuptools
+
+setuptools.setup(
+    name="bsridatta",
+    version="0.0.1",
+    author="Sri Datta Budaraju",
+    author_email="b.sridatta@gmail.com",
+    packages=setuptools.find_packages(),
+    python_requires='>=3.6'
+)

requirements.txt

@@ -0,0 +1,10 @@
+from src.callbacks.base import CallbackList, Callback
+from src.callbacks.model_checkpoint import ModelCheckpoint
+from src.callbacks.logging import Logging
+
+
+__all__ = [
+    'CallbackList',
+    'ModelCheckpoint',
+    'Logging',
+]

setup.py

@@ -0,0 +1,158 @@
+'''
+Callback inspritations from PyTorch Lightning - https://github.com/PyTorchLightning/PyTorch-Lightning
+and https://github.com/devforfu/pytorch_playground/blob/master/loop.ipynb
+'''
+
+import abc
+
+
+class Callback(abc.ABC):
+    def setup(self, **kwargs):
+        """Called before the training procedure"""
+        pass
+
+    def teardown(self, **kwargs):
+        """Called after training procedure"""
+        pass
+
+    def on_epoch_start(self, **kwargs):
+        """Called when epoch begins"""
+        pass
+
+    def on_epoch_end(self, **kwargs):
+        """Called when epoch terminates"""
+        pass
+
+    def on_train_batch_start(self, **kwargs):
+        """Called when training step begins"""
+        pass
+
+    def on_train_batch_end(self, **kwargs):
+        """Called when training step ends"""
+        pass
+
+    def on_validation_batch_start(self, **kwargs):
+        """Called when validation step begins"""
+        pass
+
+    def on_validation_batch_end(self, **kwargs):
+        """Called when validation step ends"""
+        pass
+
+    def on_test_batch_start(self, **kwargs):
+        """Called when test batch begins"""
+        pass
+
+    def on_test_batch_end(self, **kwargs):
+        """Called when test batch ends"""
+        pass
+
+    def on_train_start(self, **kwargs):
+        """Called when training loop begins"""
+        pass
+
+    def on_train_end(self, **kwargs):
+        """Called when training loop ends"""
+        pass
+
+    def on_validation_start(self, **kwargs):
+        """Called when validation loop begins"""
+        pass
+
+    def on_validation_end(self, **kwargs):
+        """Called when validation loop ends"""
+        pass
+
+    def on_test_start(self, **kwargs):
+        """Called when test loop begins"""
+        pass
+
+    def on_test_end(self, **kwargs):
+        """Called when test loop ends"""
+        pass
+
+
+class CallbackList(Callback):
+
+    def __init__(self, callbacks):
+        self.callbacks = callbacks
+
+    def setup(self, **kwargs):
+        """Called before the training procedure"""
+        for callback in self.callbacks:
+            callback.setup(**kwargs)
+
+    def teardown(self, **kwargs):
+        """Called after training procedure"""
+        for callback in self.callbacks:
+            callback.teardown(**kwargs)
+
+    def on_epoch_start(self, **kwargs):
+        """Called when epoch begins"""
+        for callback in self.callbacks:
+            callback.on_epoch_start(**kwargs)
+
+    def on_epoch_end(self, **kwargs):
+        """Called when epoch terminates"""
+        for callback in self.callbacks:
+            callback.on_epoch_end(**kwargs)
+
+    def on_train_batch_start(self, **kwargs):
+        """Called when training step begins"""
+        for callback in self.callbacks:
+            callback.on_train_batch_start(**kwargs)
+
+    def on_train_batch_end(self, **kwargs):
+        """Called when training step ends"""
+        for callback in self.callbacks:
+            callback.on_train_batch_end(**kwargs)
+
+    def on_validation_batch_start(self, **kwargs):
+        """Called when validation step begins"""
+        for callback in self.callbacks:
+            callback.on_validation_batch_start(**kwargs)
+
+    def on_validation_batch_end(self, **kwargs):
+        """Called when validation step ends"""
+        for callback in self.callbacks:
+            callback.on_validation_batch_end(**kwargs)
+
+    def on_test_batch_start(self, **kwargs):
+        """Called when test batch begins"""
+        for callback in self.callbacks:
+            callback.on_test_batch_start(**kwargs)
+
+    def on_test_batch_end(self, **kwargs):
+        """Called when test batch ends"""
+        for callback in self.callbacks:
+            callback.on_test_batch_end(**kwargs)
+
+    def on_train_start(self, **kwargs):
+        """Called when training loop begins"""
+        for callback in self.callbacks:
+            callback.on_train_start(**kwargs)
+
+    def on_train_end(self, **kwargs):
+        """Called when training loop ends"""
+        for callback in self.callbacks:
+            callback.on_train_end(**kwargs)
+
+    def on_validation_start(self, **kwargs):
+        """Called when validation loop begins"""
+        for callback in self.callbacks:
+            callback.on_validation_start(**kwargs)
+
+    def on_validation_end(self, **kwargs):
+        """Called when validation loop ends"""
+        for callback in self.callbacks:
+            callback.on_validation_end(**kwargs)
+
+    def on_test_start(self, **kwargs):
+        """Called when test loop begins"""
+        for callback in self.callbacks:
+            callback.on_test_start(**kwargs)
+
+    def on_test_end(self, **kwargs):
+        """Called when test loop ends"""
+        for callback in self.callbacks:
+            callback.on_test_end(**kwargs)

src/__init__.py

@@ -0,0 +1,52 @@
+from src.callbacks.base import Callback
+import torch
+
+
+class Logging(Callback):
+    """Logging and printing metrics"""
+
+    def setup(self, opt, model, **kwargs):
+        print(
+            f'[INFO]: Start training procedure using device: {opt.device}')
+        # log gradients and parameters of the model during training
+        if opt.use_wandb:
+            opt.logger.watch(model, log='all')
+
+    def on_train_batch_end(self, opt, batch_idx, batch, dataloader, output, l_ship, l_bbox, **kwargs):
+        batch_len = len(batch['input'])
+        dataset_len = len(dataloader.dataset)
+        n_batches = len(dataloader)
+
+        # print to console
+        print('Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.4f}\tL_ship: {:.4f}\tL_bbox: {:.4f}'.format(
+            opt.epoch, batch_idx * batch_len,
+            dataset_len, 100. * batch_idx / n_batches,
+            output, l_ship, l_bbox),
+            end='\n')
+
+        # log to wandb
+        if opt.use_wandb:
+            opt.logger.log({"train_loss": output,
+                            "l_ship": l_ship,
+                            "l_bbox": l_bbox})
+
+    def on_validation_end(self, opt, output, metrics, l_ship, l_bbox, **kwargs):
+        # print and log metrics and loss after validation epoch
+        print("Valiation - Loss: {:.4f}\tL_ship: {:.4f}\tL_bbox: {:.4f}".format(
+            output, l_ship, l_bbox), end="\t")
+
+        for k in metrics.keys():
+            print(f"{k}: {metrics[k]}", end="\t")
+            if opt.use_wandb:
+                opt.logger.log(metrics, commit=False)
+                opt.logger.log({"val_loss": output,
+                                "epoch": opt.epoch,
+                                "val_l_ship": l_ship,
+                                "val_l_bbox": l_bbox})
+        print("")
+
+    def on_epoch_end(self, opt, optimizer, **kwargs):
+        lr = optimizer.param_groups[0]['lr']
+        if opt.use_wandb:
+            opt.logger.log({f"LR": lr})
+        print("lr @ ", lr)