How to train instance segmentation with COCO dataset? about maskformer HOT 6 CLOSED

https://github.com/facebookresearch/Mask2Former supports MaskFormer as well.

from maskformer.

Thank you for your reply, can you provide me with the corresponding config file? And in the Mask2former paper, the Maskformer with R50 as the backbone is used for instance segmentation experimental results, does it additional six encoder layers as Maskformer with R50 as the backbone did for the COCO panoptic segmentation task?

from maskformer.

Hi~, I suddenly realized that I forgot to leave my email, if you can provide the config file for instance segmentation experiments with Maskformer, my email is [email protected]. Thank you!

from maskformer.

Hi @WuTao-CS ,
I am using the following config with the Mask2Fomer to run MaskFormer

BASE: Base-COCO-InstanceSegmentation.yaml
MODEL:
META_ARCHITECTURE: "MaskFormer"
SEM_SEG_HEAD:
NAME: "MaskFormerHead"
IGNORE_VALUE: 255
NUM_CLASSES: 80
LOSS_WEIGHT: 1.0
CONVS_DIM: 256
MASK_DIM: 256
NORM: "GN"
# pixel decoder
PIXEL_DECODER_NAME: "TransformerEncoderPixelDecoder"
IN_FEATURES: ["res2", "res3", "res4", "res5"]
# DEFORMABLE_TRANSFORMER_ENCODER_IN_FEATURES: ["res3", "res4", "res5"]
COMMON_STRIDE: 4
TRANSFORMER_ENC_LAYERS: 6
MASK_FORMER:
TRANSFORMER_DECODER_NAME: "StandardTransformerDecoder"
TRANSFORMER_IN_FEATURE: "transformer_encoder"
DEEP_SUPERVISION: True
NO_OBJECT_WEIGHT: 0.1
DICE_WEIGHT: 1.0
MASK_WEIGHT: 20.0
CLASS_WEIGHT: 2.0
HIDDEN_DIM: 256
NUM_OBJECT_QUERIES: 100
NHEADS: 8
DROPOUT: 0.1
DIM_FEEDFORWARD: 2048
ENC_LAYERS: 0
DEC_LAYERS: 6
PRE_NORM: False
# ENFORCE_INPUT_PROJ: False
SIZE_DIVISIBILITY: 0
# DEC_LAYERS: 10 # 9 decoder layers, add one for the loss on learnable query
# TRAIN_NUM_POINTS: 12544
# OVERSAMPLE_RATIO: 3.0
# IMPORTANCE_SAMPLE_RATIO: 0.75
TEST:
SEMANTIC_ON: False
INSTANCE_ON: True
PANOPTIC_ON: False
OVERLAP_THRESHOLD: 0.8
OBJECT_MASK_THRESHOLD: 0.8
SOLVER:
AMP:
ENABLED: TRUE
INPUT:
MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800)
CROP:
ENABLED: True
TYPE: "absolute_range"
SIZE: (384, 600)

from maskformer.

Hi @rmanzanedo ，thanks for your help！

from maskformer.

Hey @WuTao-CS, Did you successfully trained it for image segmentation. What was your experience?

from maskformer.