add EPIC100DatasetAccess

kale/loaddata/video_access.py

@@ -11,12 +11,15 @@
 import os
 from copy import deepcopy
 from enum import Enum
+from pathlib import Path
 
+import pandas as pd
 import torch
 
 import kale.prepdata.video_transform as video_transform
 from kale.loaddata.dataset_access import DatasetAccess
 from kale.loaddata.video_datasets import BasicVideoDataset, EPIC
+from kale.loaddata.videos import VideoFrameDataset
 
 
 def get_image_modality(image_modality):
@@ -213,13 +216,14 @@ class VideoDatasetAccess(DatasetAccess):
  """
 
  def __init__(
- self, data_path, train_list, test_list, image_modality, frames_per_segment, n_classes, transform_kind, seed
+  self, data_path, train_list, test_list, image_modality, num_segments, frames_per_segment, n_classes, transform_kind, seed
  ):
  super().__init__(n_classes)
  self._data_path = data_path
  self._train_list = train_list
  self._test_list = test_list
  self._image_modality = image_modality
+ self._num_segments = num_segments
  self._frames_per_segment = frames_per_segment
  self._transform = video_transform.get_transform(transform_kind, self._image_modality)
  self._seed = seed
@@ -370,3 +374,75 @@ def get_test(self):
  dataset_split="test",
  n_classes=self._n_classes,
  )
+
+
+class EPIC100DatasetAccess(VideoDatasetAccess):
+ """EPIC-100 video feature data loader"""
+
+ def __init__(
+ self,
+ domain,
+ data_path,
+ train_list,
+ test_list,
+ image_modality,
+ num_segments,
+ frames_per_segment,
+ n_classes,
+ transform,
+ seed,
+ input_type,
+ ):
+ super(EPIC100DatasetAccess, self).__init__(
+ data_path,
+ train_list,
+ test_list,
+ image_modality,
+ num_segments,
+ frames_per_segment,
+ n_classes,
+ transform,
+ seed,
+ )
+ self._input_type = input_type
+ self._domain = domain
+ self._num_train_dataload = len(pd.read_pickle(self._train_list).index)
+ self._num_test_dataload = len(pd.read_pickle(self._test_list).index)
+
+ def get_train(self):
+ return VideoFrameDataset(
+ root_path=Path(self._data_path, self._input_type, "{}_val.pkl".format(self._domain)),
+ # Uncomment to run on train subset for EPIC 2021 challenge
+ # data_path=Path(self._data_path, self._input_type, "{}_train.pkl".format(self._domain)),
+ annotationfile_path=self._train_list,
+ total_segments=25,
+ num_segments=self._num_segments, # 1
+ frames_per_segment=self._frames_per_segment, # 1
+ image_modality=self._image_modality,
+ imagefile_template="img_{:05d}.t7"
+ if self._image_modality in ["RGB", "RGBDiff", "RGBDiff2", "RGBDiffplus"]
+ else self._input_type + "{}_{:05d}.t7",
+ random_shift=False,
+ test_mode=True,
+ input_type="feature",
+ num_data_load=self._num_train_dataload,
+ )
+
+ def get_test(self):
+ return VideoFrameDataset(
+ root_path=Path(self._data_path, self._input_type, "{}_val.pkl".format(self._domain)),
+ # Uncomment to run on test subset for EPIC 2021 challenge
+ # data_path=Path(self._data_path, self._input_type, "{}_test.pkl".format(self._domain)),
+ annotationfile_path=self._test_list,
+ total_segments=25,
+ num_segments=self._num_segments,
+ frames_per_segment=self._frames_per_segment,
+ image_modality=self._image_modality,
+ imagefile_template="img_{:05d}.t7"
+ if self._image_modality in ["RGB", "RGBDiff", "RGBDiff2", "RGBDiffplus"]
+ else self._input_type + "{}_{:05d}.t7",
+ random_shift=True,
+ test_mode=True,
+ input_type="feature",
+ num_data_load=self._num_test_dataload,
+ )