How to run new Baselines (LazyConfig) with Colab Demo

[harezhh]

Dear Community,

I'm beginner in the field of Object Detection and I try my first project. I was able to detect with your Colab Demo the object of my dataset with Mask R-CNN. Thanks for that ! :D

Then I noticed that there are new Baselines for Mask R-CNN with better mAP-results. Unfortunately I'm still not able to run the same Colab Demo with the new Baselines. I read the documentary and other similar problems but without success. Thats the reason I'm asking here. How should I modify this peace of code to be able running training with new Baselines?

This is the Demo I used.

`from detectron2.engine import DefaultTrainer
cfg = get_cfg()
cfg.merge_from_file(model_zoo.get_config_file("COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml"))
cfg.DATASETS.TRAIN = ("balloon_train",)
cfg.DATASETS.TEST = ()
cfg.DATALOADER.NUM_WORKERS = 2
cfg.MODEL.WEIGHTS = model_zoo.get_checkpoint_url("COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml")
cfg.SOLVER.IMS_PER_BATCH = 2
cfg.SOLVER.BASE_LR = 0.00025
cfg.SOLVER.MAX_ITER = 300
cfg.SOLVER.STEPS = []
cfg.MODEL.ROI_HEADS.BATCH_SIZE_PER_IMAGE = 128
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 1

os.makedirs(cfg.OUTPUT_DIR, exist_ok=True)
trainer = DefaultTrainer(cfg)
trainer.resume_or_load(resume=False)
trainer.train()`

Know I tried to change if according to LazyConfig-System. But at this point I'm not able to let it run.

from detectron2.config import instantiate from detectron2.model_zoo import get_config model_name="new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_100ep_LSJ.py" cfg = LazyConfig.load(model_zoo.get_config_file(model_name)) cfg.dataloader.train.dataset.names = 'my_dataset_train' cfg.dataloader.test.dataset.names = 'my_dataset_test' cfg.MAX_ITER = 3000 cfg.NUM_CLASSES = 1 cfg.lr = 0.01 trainer = DefaultTrainer(cfg) trainer.resume_or_load(resume=False) trainer.train()

I'm greatfull for any help. Thanks.

[FeriBolour]

I learned everything I need from here:
#3225