Merge pull request #45 from Purg/dev/random-transform-aug

Random transform aug
PTG-Kitware · Nov 14, 2024 · 3c5cb82 · 3c5cb82
2 parents d904388 + 25329a4
commit 3c5cb82
Show file tree

Hide file tree

Showing 5 changed files with 400 additions and 18 deletions.
diff --git a/configs/experiment/m2/feat_locsconfs.yaml b/configs/experiment/m2/feat_locsconfs.yaml
@@ -98,6 +98,8 @@ data:
           dets_throughput_std: 0.2
           pose_throughput_std: 0.2
           fixed_pattern: false
+        - _target_: tcn_hpl.data.frame_data_aug.rotate_scale_translate_jitter.FrameDataRotateScaleTranslateJitter
+          # Using default parameters to start
   val_dataset:
     # Augmentations on windows of frame data before performing vectorization.
     # Sharing transform with training dataset as it is only the drop-out aug to
@@ -106,13 +108,13 @@ data:
       transforms:
         - _target_: tcn_hpl.data.frame_data_aug.window_frame_dropout.DropoutFrameDataTransform
           # Mirror training hparams, except used fixed patterns.
-          frame_rate: ${data.train_dataset.frame_rate}
-          dets_throughput_mean: ${data.train_dataset.dets_throughput_mean}
-          pose_throughput_mean: ${data.train_dataset.pose_throughput_mean}
-          dets_latency: ${data.train_dataset.dets_latency}
-          pose_latency: ${data.train_dataset.pose_latency}
-          dets_throughput_std: ${data.train_dataset.dets_throughput_std}
-          pose_throughput_std: ${data.train_dataset.pose_throughput_std}
+          frame_rate: ${data.target_framerate}
+          dets_throughput_mean: 14.5
+          pose_throughput_mean: 10
+          dets_latency: 0
+          pose_latency: 0.1
+          dets_throughput_std: 0.2
+          pose_throughput_std: 0.2
           fixed_pattern: true
   # Test dataset usually configured the same as val, unless there is some
   # different set of transforms that should be used during test/prediction.

diff --git a/configs/experiment/r18/feat_locsconfs.yaml b/configs/experiment/r18/feat_locsconfs.yaml
@@ -92,6 +92,8 @@ data:
           dets_throughput_std: 0.2
           pose_throughput_std: 0.2
           fixed_pattern: false
+        - _target_: tcn_hpl.data.frame_data_aug.rotate_scale_translate_jitter.FrameDataRotateScaleTranslateJitter
+          # Using default parameters to start
   val_dataset:
     # Augmentations on windows of frame data before performing vectorization.
     # Sharing transform with training dataset as it is only the drop-out aug to
@@ -100,13 +102,13 @@ data:
       transforms:
         - _target_: tcn_hpl.data.frame_data_aug.window_frame_dropout.DropoutFrameDataTransform
           # Mirror training hparams, except used fixed patterns.
-          frame_rate: ${data.train_dataset.frame_rate}
-          dets_throughput_mean: ${data.train_dataset.dets_throughput_mean}
-          pose_throughput_mean: ${data.train_dataset.pose_throughput_mean}
-          dets_latency: ${data.train_dataset.dets_latency}
-          pose_latency: ${data.train_dataset.pose_latency}
-          dets_throughput_std: ${data.train_dataset.dets_throughput_std}
-          pose_throughput_std: ${data.train_dataset.pose_throughput_std}
+          frame_rate: ${data.target_framerate}
+          dets_throughput_mean: 14.5
+          pose_throughput_mean: 10
+          dets_latency: 0
+          pose_latency: 0.1
+          dets_throughput_std: 0.2
+          pose_throughput_std: 0.2
           fixed_pattern: true
   # Test dataset usually configured the same as val, unless there is some
   # different set of transforms that should be used during test/prediction.

diff --git a/tcn_hpl/data/frame_data.py b/tcn_hpl/data/frame_data.py
@@ -73,6 +73,11 @@ class FrameData:
     """
     Structure composing information and correlated analytic results for a
     single image frame.
+
+    Args:
+        object_detections: Optional initial frame object detections.
+        poses: Optional initial pose estimations
+        size: Frame pixel dimensions in (width, height) format (integer).
     """
 
     # Object detection results for this frame.