fix imports

gluoncv/auto/estimators/center_net/center_net.py

@@ -210,8 +210,7 @@ def _evaluate(self, val_data):
         mx.nd.waitall()
         self.net.hybridize()
         if not isinstance(val_data, gluon.data.DataLoader):
-            from ...tasks.dataset import ObjectDetectionDataset
-            if isinstance(val_data, ObjectDetectionDataset):
+            if hasattr(val_data, 'to_mxnet'):
                 val_data = val_data.to_mxnet()
             val_batchify_fn = Tuple(Stack(), Pad(pad_val=-1))
             width, height = self._cfg.center_net.data_shape

gluoncv/auto/estimators/faster_rcnn/default.py

@@ -161,6 +161,8 @@ class TrainCfg:
 
 @dataclass
 class ValidCfg:
+    # Batch size during training
+    batch_size : int = 1
     # Filter top proposals before NMS in testing of RPN.
     rpn_test_pre_nms : int = 6000
     # Return top proposal results after NMS in testing of RPN.

gluoncv/auto/estimators/faster_rcnn/faster_rcnn.py

@@ -8,7 +8,7 @@
 import mxnet as mx
 from mxnet import gluon
 
-from ....data.batchify import Tuple
+from ....data.batchify import Tuple, Append
 from ....data.transforms.presets.rcnn import load_test, transform_test
 from ....data.transforms.presets.rcnn import FasterRCNNDefaultTrainTransform, FasterRCNNDefaultValTransform
 from ....model_zoo import get_model
@@ -219,16 +219,15 @@ def _evaluate(self, val_data):
         """Evaluate on validation dataset."""
         clipper = BBoxClipToImage()
         if not isinstance(val_data, gluon.data.DataLoader):
-            from ...tasks.dataset import ObjectDetectionDataset
-            if isinstance(val_data, ObjectDetectionDataset):
+            if hasattr(val_data, 'to_mxnet'):
                 val_data = val_data.to_mxnet()
             val_bfn = Tuple(*[Append() for _ in range(3)])
-            short = net.short[-1] if isinstance(net.short, (tuple, list)) else net.short
+            short = self.net.short[-1] if isinstance(self.net.short, (tuple, list)) else self.net.short
             # validation use 1 sample per device
             val_data = gluon.data.DataLoader(
-                val_data.transform(FasterRcnnDefaultValTransform(short, net.max_size)),
+                val_data.transform(FasterRCNNDefaultValTransform(short, self.net.max_size)),
                 len(self.ctx), False, batchify_fn=val_bfn, last_batch='keep',
-                num_workers=self._cfg.valid.num_workers)
+                num_workers=self._cfg.num_workers)
         if self._cfg.valid.metric == 'voc07':
             eval_metric = VOC07MApMetric(iou_thresh=self._cfg.valid.iou_thresh, class_names=self.classes)
         elif self._cfg.valid.metric == 'voc':

gluoncv/auto/estimators/image_classification/image_classification.py

@@ -316,10 +316,11 @@ def _evaluate(self, val_data):
         acc_top1 = mx.metric.Accuracy()
         acc_top5 = mx.metric.TopKAccuracy(min(5, self.num_class))
 
-        if not isinstance(val_data, (gluon.data.DataLoader, mx.io.ImageRecordIter)):
-            from ...tasks.dataset import ImageClassificationDataset
-            if isinstance(val_data, ImageClassificationDataset):
+        if not isinstance(val_data, (gluon.data.DataLoader, mx.io.MXDataIter)):
+            if hasattr(val_data, 'to_mxnet'):
                 val_data = val_data.to_mxnet()
+            resize = int(math.ceil(self.input_size / self._cfg.train.crop_ratio))
+            normalize = transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
             transform_test = transforms.Compose([
                 transforms.Resize(resize, keep_ratio=True),
                 transforms.CenterCrop(self.input_size),

gluoncv/auto/estimators/image_classification/utils.py

@@ -78,7 +78,6 @@ def get_data_loader(data_dir, batch_size, num_workers, input_size, crop_ratio, t
     normalize = transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
     jitter_param = 0.4
     lighting_param = 0.1
-    input_size = input_size
     crop_ratio = crop_ratio if crop_ratio > 0 else 0.875
     resize = int(math.ceil(input_size / crop_ratio))
 

gluoncv/auto/estimators/ssd/default.py

@@ -60,6 +60,8 @@ class TrainCfg:
 
 @dataclass
 class ValidCfg:
+    # Batch size during training
+    batch_size : int = 16
     # Epoch interval for validation
     val_interval : int = 1
     # metric, 'voc', 'voc07'

gluoncv/auto/estimators/ssd/ssd.py

@@ -17,6 +17,7 @@
 from ....utils.metrics.voc_detection import VOC07MApMetric, VOCMApMetric
 from ....model_zoo import get_model
 from ....model_zoo import custom_ssd
+from ....data.transforms.presets.ssd import SSDDefaultValTransform
 from ....data.transforms.presets.ssd import load_test, transform_test
 from ....loss import SSDMultiBoxLoss
 from .utils import _get_dataloader, _get_dali_dataloader
@@ -204,14 +205,13 @@ def _train_loop(self, train_data, val_data, train_eval_data):
     def _evaluate(self, val_data):
         """Evaluate on validation dataset."""
         if not isinstance(val_data, gluon.data.DataLoader):
-            from ...tasks.dataset import ObjectDetectionDataset
-            if isinstance(val_data, ObjectDetectionDataset):
+            if hasattr(val_data, 'to_mxnet'):
                 val_data = val_data.to_mxnet()
             val_batchify_fn = Tuple(Stack(), Pad(pad_val=-1))
             val_data = gluon.data.DataLoader(
-                val_data.transform(SSDDefaultValTransform(width, height)),
+                val_data.transform(SSDDefaultValTransform(self._cfg.ssd.data_shape, self._cfg.ssd.data_shape)),
                 self._cfg.valid.batch_size, False, batchify_fn=val_batchify_fn, last_batch='keep',
-                num_workers=self._cfg.valid.num_workers)
+                num_workers=self._cfg.num_workers)
         if self._cfg.valid.metric == 'voc07':
             eval_metric = VOC07MApMetric(iou_thresh=self._cfg.valid.iou_thresh, class_names=self.classes)
         elif self._cfg.valid.metric == 'voc':

gluoncv/auto/estimators/yolo/default.py

@@ -78,6 +78,8 @@ class TrainCfg:
 
 @dataclass
 class ValidCfg:
+    # Batch size during training
+    batch_size : int = 16
     # Epoch interval for validation, increase the number
     # will reduce the training time if validation is slow.
     val_interval : int = 1

gluoncv/auto/estimators/yolo/yolo.py

@@ -12,6 +12,7 @@
 
 from .... import utils as gutils
 from ....data.batchify import Tuple, Stack, Pad
+from ....data.transforms.presets.yolo import YOLO3DefaultValTransform
 from ....data.transforms.presets.yolo import load_test, transform_test
 from ....model_zoo import get_model
 from ....model_zoo import custom_yolov3
@@ -202,14 +203,13 @@ def _train_loop(self, train_data, val_data, train_eval_data):
     def _evaluate(self, val_data):
         """Evaluate the current model on dataset."""
         if not isinstance(val_data, gluon.data.DataLoader):
-            from ...tasks.dataset import ObjectDetectionDataset
-            if isinstance(val_data, ObjectDetectionDataset):
+            if hasattr(val_data, 'to_mxnet'):
                 val_data = val_data.to_mxnet()
             val_batchify_fn = Tuple(Stack(), Pad(pad_val=-1))
             val_data = gluon.data.DataLoader(
-                val_data.transform(SSDDefaultValTransform(width, height)),
+                val_data.transform(YOLO3DefaultValTransform(self._cfg.yolo3.data_shape, self._cfg.yolo3.data_shape)),
                 self._cfg.valid.batch_size, False, batchify_fn=val_batchify_fn, last_batch='keep',
-                num_workers=self._cfg.valid.num_workers)
+                num_workers=self._cfg.num_workers)
         if self._cfg.valid.metric == 'voc07':
             eval_metric = VOC07MApMetric(iou_thresh=self._cfg.valid.iou_thresh, class_names=self.classes)
         elif self._cfg.valid.metric == 'voc':

tests/auto/test_auto_estimators.py

@@ -31,6 +31,7 @@ def test_image_classification_estimator():
     res = est.fit(IMAGE_CLASS_DATASET)
     assert res.get('valid_acc', 0) > 0
     test_result = est.predict(IMAGE_CLASS_TEST)
+    evaluate_result = est.evaluate(IMAGE_CLASS_TEST)
 
 def test_center_net_estimator():
     from gluoncv.auto.estimators import CenterNetEstimator
@@ -39,6 +40,7 @@ def test_center_net_estimator():
     assert res.get('valid_map', 0) > 0
     _, _, test_data = OBJECT_DETCTION_DATASET.random_split()
     test_result = est.predict(test_data)
+    evaluate_result = est.evaluate(test_data)
 
 def test_ssd_estimator():
     from gluoncv.auto.estimators import SSDEstimator
@@ -47,6 +49,7 @@ def test_ssd_estimator():
     assert res.get('valid_map', 0) > 0
     _, _, test_data = OBJECT_DETCTION_DATASET.random_split()
     test_result = est.predict(test_data)
+    evaluate_result = est.evaluate(test_data)
 
 def test_yolo3_estimator():
     from gluoncv.auto.estimators import YOLOv3Estimator
@@ -55,6 +58,7 @@ def test_yolo3_estimator():
     assert res.get('valid_map', 0) > 0
     _, _, test_data = OBJECT_DETCTION_DATASET.random_split()
     test_result = est.predict(test_data)
+    evaluate_result = est.evaluate(test_data)
 
 def test_frcnn_estimator():
     from gluoncv.auto.estimators import FasterRCNNEstimator
@@ -63,6 +67,7 @@ def test_frcnn_estimator():
     assert res.get('valid_map', 0) > 0
     _, _, test_data = OBJECT_DETCTION_DATASET.random_split()
     test_result = est.predict(test_data)
+    evaluate_result = est.evaluate(test_data)
 
 if __name__ == '__main__':
     import nose