[Feature] Support MaskDINO cityscapes semantic seg

projects/maskdino/configs/data/cityscapes_semantic_seg.py

@@ -0,0 +1,58 @@
+from omegaconf import OmegaConf
+
+import detectron2.data.transforms as T
+from detectron2.config import LazyCall as L
+from detectron2.data import (
+    build_detection_test_loader,
+    build_detection_train_loader,
+    get_detection_dataset_dicts,
+)
+from detectron2.data.dataset_mapper import DatasetMapper
+from detectron2.evaluation import CityscapesSemSegEvaluator
+from detectron2.data import MetadataCatalog
+
+# from detrex.data import DetrDatasetMapper
+# from projects.maskDINO.data.dataset_mappers.coco_instance_lsj_aug_dataset_mapper import COCOInstanceLSJDatasetMapper, build_transform_gen
+from detrex.data.dataset_mappers.mask_former_semantic_dataset_mapper import build_transform_gen, MaskFormerSemanticDatasetMapper
+
+dataloader = OmegaConf.create()
+
+dataloader.train = L(build_detection_train_loader)(
+    dataset=L(get_detection_dataset_dicts)(names="cityscapes_fine_sem_seg_train"),
+    mapper=L(MaskFormerSemanticDatasetMapper)(
+        augmentation=L(build_transform_gen)(
+            min_size_train=[int(x * 0.1 * 1024) for x in range(5, 21)],
+            max_size_train=4096,
+            min_size_train_sampling='choice',
+            enabled_crop=True,
+            crop_params=dict(crop_type='absolute', crop_size=(512, 1024), single_category_max_area=1.0),
+            color_aug_ssd=True,
+            img_format='RGB',
+        ),
+        meta=MetadataCatalog.get("cityscapes_fine_sem_seg_train"),
+        size_divisibility=-1,
+        is_train=True,
+        image_format="RGB",
+    ),
+    total_batch_size=16,
+    num_workers=4,
+)
+
+dataloader.test = L(build_detection_test_loader)(
+    dataset=L(get_detection_dataset_dicts)(names="cityscapes_fine_sem_seg_val", filter_empty=False),
+    mapper=L(DatasetMapper)(
+        augmentation=[
+            L(T.ResizeShortestEdge)(
+                short_edge_length=1024,
+                max_size=4096,
+            ),
+        ],
+        is_train=False,
+        image_format="RGB",
+    ),
+    num_workers=4,
+)
+
+dataloader.evaluator = L(CityscapesSemSegEvaluator)(
+    dataset_name="${..test.dataset.names}",
+)

projects/maskdino/configs/maskdino_r50_cityscapes_semantic_seg_50ep.py

@@ -0,0 +1,67 @@
+from detrex.config import get_config
+from .models.maskdino_r50 import model
+from .data.cityscapes_semantic_seg import dataloader
+
+from fvcore.common.param_scheduler import MultiStepParamScheduler
+from detectron2.config import LazyCall as L
+from detectron2.solver import WarmupParamScheduler
+
+model.semantic_on=True
+model.instance_on=False
+model.panoptic_on=False
+
+train = get_config("common/train.py").train
+# max training iterations
+train.max_iter = 368750
+# warmup lr scheduler
+lr_multiplier = L(WarmupParamScheduler)(
+    scheduler=L(MultiStepParamScheduler)(
+        values=[1.0, 0.1],
+        milestones=[327778, 355092],
+    ),
+    warmup_length=10 / train.max_iter,
+    warmup_factor=1.0,
+)
+
+optimizer = get_config("common/optim.py").AdamW
+# lr_multiplier = get_config("common/coco_schedule.py").lr_multiplier_50ep
+
+# initialize checkpoint to be loaded
+train.init_checkpoint = "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
+train.output_dir = "./output/dab_detr_r50_50ep"
+
+
+# run evaluation every 5000 iters
+train.eval_period = 5000
+
+# log training infomation every 20 iters
+train.log_period = 20
+
+# save checkpoint every 5000 iters
+train.checkpointer.period = 5000
+
+# gradient clipping for training
+train.clip_grad.enabled = True
+train.clip_grad.params.max_norm = 0.01
+train.clip_grad.params.norm_type = 2
+
+# set training devices
+train.device = "cuda"
+
+
+# modify optimizer config
+optimizer.lr = 1e-4
+optimizer.betas = (0.9, 0.999)
+optimizer.weight_decay = 1e-4
+optimizer.params.lr_factor_func = lambda module_name: 0.1 if "backbone" in module_name else 1
+
+# # modify dataloader config
+dataloader.train.num_workers = 16
+#
+# # please notice that this is total batch size.
+# # surpose you're using 4 gpus for training and the batch size for
+# # each gpu is 16/4 = 4
+dataloader.train.total_batch_size = 16
+
+# dump the testing results into output_dir for visualization
+dataloader.evaluator.output_dir = train.output_dir