code corresponding to sampling process and text list generation during training about oneformer

[praeclarumjj3]

Hi, @yq790, thanks for your interest in OneFormer.

We randomly sample the task inside the corresponding DatasetMapper. the probabilities are defined inside the config files.

OneFormer/configs/ade20k/Base-ADE20K-UnifiedSegmentation.yaml

Lines 55 to 57 in 56799ef

	TASK_PROB:
	SEMANTIC: 0.33
	INSTANCE: 0.66

OneFormer/oneformer/data/dataset_mappers/oneformer_unified_dataset_mapper.py

Lines 351 to 366 in 56799ef

	prob_task = np.random.uniform(0,1.)
	num_class_obj = {}
	for name in self.class_names:
	num_class_obj[name] = 0
	if prob_task < self.semantic_prob:
	task = "The task is semantic"
	instances, text, sem_seg = self._get_semantic_dict(pan_seg_gt, image_shape, segments_info, num_class_obj)
	elif prob_task < self.instance_prob:
	task = "The task is instance"
	instances, text, sem_seg = self._get_instance_dict(pan_seg_gt, image_shape, segments_info, num_class_obj)
	else:
	task = "The task is panoptic"
	instances, text, sem_seg = self._get_panoptic_dict(pan_seg_gt, image_shape, segments_info, num_class_obj)

[praeclarumjj3]

Closing, feel free to reopen.