search

mrlooi / rotated_maskrcnn

Rotated Mask R-CNN: From Bounding Boxes to Rotated Bounding Boxes
MIT License
347 stars 62 forks source link

Loss is too high #42

Open psinha30 opened 3 years ago

psinha30 commented 3 years ago

❓ Questions and Help

Hi, I was trying to train the mrcnn on a custom dataset but seems the loss is very high. Is this correct and also I dont see any signs of convergence. image

the config file: MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50" ROTATED: True BACKBONE: CONV_BODY: "R-50-FPN" RESNETS: BACKBONE_OUT_CHANNELS: 256 RPN: USE_FPN: True ANCHOR_STRIDE: (4, 8, 16, 32, 64) PRE_NMS_TOP_N_TRAIN: 2000 PRE_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TEST: 1000 FPN_POST_NMS_TOP_N_TEST: 1000

STRADDLE_THRESH: -1
ANCHOR_ANGLES: (-90, -60, -30)

BBOX_REG_WEIGHTS: (1.0, 1.0, 1.0, 1.0, 1.0)

ROI_HEADS: USE_FPN: True

# weights on (dx, dy, dw, dh, dtheta) for normalizing rotated rect regression targets
BBOX_REG_WEIGHTS: (10.0, 10.0, 5.0, 5.0, 1.0)

USE_SOFT_NMS: True
SOFT_NMS:
  METHOD: 1

ROI_BOX_HEAD: POOLER_RESOLUTION: 7 POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) POOLER_SAMPLING_RATIO: 2 FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" PREDICTOR: "FPNPredictor" ROI_MASK_HEAD: POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" PREDICTOR: "MaskRCNNC4Predictor" POOLER_RESOLUTION: 14 POOLER_SAMPLING_RATIO: 2 RESOLUTION: 28 SHARE_BOX_FEATURE_EXTRACTOR: False MASK_ON: True

MASKIOU_ON: True ROI_MASKIOU_HEAD: USE_NMS: True DATASETS: TRAIN: ("cocodataset_train","cocodataset_val") TEST: ("cocodataset_val",) DATALOADER: SIZE_DIVISIBILITY: 32 SOLVER: BASE_LR: 0.0005 WEIGHT_DECAY: 0.0001 STEPS: (60000, 80000) MAX_ITER: 100000

OUTPUT_DIR: "checkpoints/rotated/mscoco_msrcnn"

psinha30 commented 3 years ago

EXAMPLE TRAINING IMAGE image