writing tests for empty frames, we will need to bump torchvision version

deepforest/dataset.py

@@ -70,6 +70,7 @@ def __getitem__(self, idx):
         image = image.astype("float32")
 
         if self.train:
+
             # select annotations
             image_annotations = self.annotations[self.annotations.image_path ==
                                                  self.image_names[idx]]

dev_requirements.txt

@@ -21,7 +21,7 @@ sphinx_rtd_theme
 sphinx_markdown_tables
 tqdm
 twine
-torchvision
+torchvision >= 0.9.0
 xmltodict
 comet_ml
 imagecodecs

environment.yml

@@ -2,7 +2,6 @@ name: DeepForest
 channels:
   - pytorch
   - conda-forge
-  - defaults
 dependencies:
   - sphinx
   - recommonmark

tests/test_main.py

@@ -85,9 +85,15 @@ def test_use_bird_release(m):
     boxes = m.predict_image(path=imgpath)
     assert not boxes.empty
 
-def test_train_single(m):
+def test_train_empty(m, tmpdir):
+    empty_csv = pd.DataFrame({"image_path":"OSBS_029.png","xmin":[0],"xmax":[0],"ymin":[0],"ymax":[0],"label":["Tree"]})
+    empty_csv.to_csv("{}/empty.csv".format(tmpdir))
+    m.config["train"]["csv_file"] = "{}/empty.csv".format(tmpdir)
     m.trainer.fit(m)
 
+def test_train_single(m):
+    m.trainer.fit(m)
+
 def test_train_multi(two_class_m):
     two_class_m.trainer.fit(two_class_m)
 

tests/test_model.py

@@ -3,6 +3,23 @@
 import pytest
 import numpy as np
 import torch
+import torchvision
+import os
+
+os.environ['KMP_DUPLICATE_LIB_OK']='True'
+
+#Empty tester from https://github.com/datumbox/vision/blob/06ebee1a9f10c76d8ac5768fd578362dd5ace6e9/test/test_models_detection_negative_samples.py#L14
+def _make_empty_sample():
+    images = [torch.rand((3, 100, 100), dtype=torch.float32)]
+    boxes = torch.zeros((0, 4), dtype=torch.float32)
+    negative_target = {"boxes": boxes,
+                       "labels": torch.zeros(0, dtype=torch.int64),
+                       "image_id": 4,
+                       "area": (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0]),
+                       "iscrowd": torch.zeros((0,), dtype=torch.int64)}
+
+    targets = [negative_target]
+    return images, targets
 
 def test_load_backbone():
     retinanet = model.load_backbone()
@@ -17,3 +34,17 @@ def test_create_model(num_classes):
     retinanet_model.eval()
     x = [torch.rand(3, 300, 400), torch.rand(3, 500, 400)]
     predictions = retinanet_model(x)    
+
+def test_forward_empty():
+    retinanet_model = model.create_model(num_classes=2,nms_thresh=0.1, score_thresh=0.2)
+    image, targets = _make_empty_sample()
+    loss = retinanet_model(image, targets)
+
+def test_forward_negative_sample_retinanet(self):
+    model = torchvision.models.detection.retinanet_resnet50_fpn(
+        num_classes=2, min_size=100, max_size=100)
+
+    images, targets = self._make_empty_sample()
+    loss_dict = model(images, targets)
+
+    self.assertEqual(loss_dict["bbox_regression"], torch.tensor(0.))