yiruoyanyu / hANMCL

复现该模型,更加详细的注释与数据集自定义,以及自己修改模块
0 stars 0 forks source link

for fine-tune #1

Open dogdog258 opened 7 months ago

dogdog258 commented 7 months ago

When i use this sentense "python train.py --dataset coco_ft --flip --net DAnA --lr 0.001 --lr_decay_step 12 --bs 2 --epochs 12 --disp_interval 20 --savedir models/{SEED}{SHOTS} --way 2 --shot {TRAIN SHOTS} --seed seed{SEED} --shots {SHOTS}shots --r" to fine-tune。It returns to me "usage: train.py [-h] [--dataset DATASET] [--net NET] [--flip] [--o OPTIMIZER] [--lr LR] [--lr_decay_step LR_DECAY_STEP] [--lr_decay_gamma LR_DECAY_GAMMA] [--nw NUM_WORKERS] [--ls] [--mGPUs] [--bs BATCH_SIZE] [--start_epoch START_EPOCH] [--epochs MAX_EPOCHS] [--disp_interval DISP_INTERVAL] [--save_dir SAVE_DIR] [--ascale ASCALE] [--eval] [--onc OLD_N_CLASSES] [--eval_dir EVAL_DIR] [--fs] [--way WAY] [--shot SHOT] [--sup_dir SUP_DIR] [--r] [--load_dir LOAD_DIR] [--checkepoch CHECKEPOCH] [--checkpoint CHECKPOINT] [--dlog] [--imlog] [--sup FT_SUP] [--seed SEED] [--shots SHOTS] train.py: error: argument --shot: invalid int value: '{TRAIN'". How to slove it.

yiruoyanyu commented 7 months ago

you should make --shot {TRAIN SHOTS} the TRAIN-SHOTS to an int ; like 10; please modify your bash code but i dont suggest you use this project ,the structure of this project is not clear。 you can try VFA to learn few shot object detection ,the link is https://github.com/csuhan/VFA