ChatGPT-ImageCaptioner / configs /Detic_LbaseI_CLIP_R5021k_640b64_4x_ft4x_predicted.yaml
taesiri's picture
Duplicate from taesiri/DeticChatGPT
f97cf44
raw
history blame contribute delete
724 Bytes
_BASE_: "Base-C2_L_R5021k_640b64_4x.yaml"
MODEL:
ROI_BOX_HEAD:
USE_ZEROSHOT_CLS: True
IMAGE_LABEL_LOSS: 'max_score'
WEIGHTS: "models/BoxSup-C2_Lbase_CLIP_R5021k_640b64_4x.pth"
SOLVER:
MAX_ITER: 90000
IMS_PER_BATCH: 64
BASE_LR: 0.0002
WARMUP_ITERS: 1000
WARMUP_FACTOR: 0.001
DATASETS:
TRAIN: ("lvis_v1_train_norare","imagenet_lvis_v1")
DATALOADER:
SAMPLER_TRAIN: "MultiDatasetSampler"
DATASET_RATIO: [1, 4]
USE_DIFF_BS_SIZE: True
DATASET_BS: [8, 32]
DATASET_INPUT_SIZE: [640, 320]
USE_RFS: [True, False]
DATASET_INPUT_SCALE: [[0.1, 2.0], [0.5, 1.5]]
FILTER_EMPTY_ANNOTATIONS: False
MULTI_DATASET_GROUPING: True
DATASET_ANN: ['box', 'image']
NUM_WORKERS: 8
WITH_IMAGE_LABELS: True