Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Base40 results not good #13

Open
YuQianzi opened this issue Sep 25, 2022 · 1 comment
Open

Base40 results not good #13

YuQianzi opened this issue Sep 25, 2022 · 1 comment

Comments

@YuQianzi
Copy link

I cannnot reopen issue #12.
Actually, when I only train the first 40 base classes using the configs in the file "warp_faster_rcnn_R_50_C4_1x.yaml", the AP50 was about 30%. I think it is not reasonable.

@YuQianzi
Copy link
Author

@JosephKJ :Hello! My configs is shown below:

BASE: "../Base-RCNN-C4.yaml"
MODEL:
WEIGHTS: "/sharefs/wwen/incremental/iOD/IOD/configs/MSRA-R-50.pkl"
MASK_ON: False
RESNETS:
DEPTH: 50
ROI_HEADS:
NUM_CLASSES: 80
LEARN_INCREMENTALLY: True
TRAIN_ON_BASE_CLASSES: True
NUM_BASE_CLASSES: 40
NUM_NOVEL_CLASSES: 40
NMS_THRESH_TEST: 0.4
DATASETS:
TRAIN: ("coco_2014_train",)
TEST: ("coco_2014_val",)

OUTPUT_DIR: ./output/coco_base_40
SEED: 3074309
VERSION: 2

SOLVER:
IMS_PER_BATCH: 16
BASE_LR: 0.02
STEPS: (140000, 160000)
MAX_ITER: 180000

WG:
ENABLE: False
TRAIN_WARP_AT_ITR_NO: 20
WARP_LAYERS: ("module.roi_heads.res5.2.conv3.weight",)
NUM_FEATURES_PER_CLASS: 100
NUM_IMAGES_PER_CLASS: 10
BATCH_SIZE: 2
USE_FEATURE_STORE: True
IMAGE_STORE_LOC: '/40_image_store.pth'

DISTILL:
ENABLE: False
BACKBONE: True
RPN: False
ROI_HEADS: True
ONLY_FG_ROIS: False
LOSS_WEIGHT: 0.2

the result of AP50 of the base 40 classes is 30.4642, which made me confused.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant