A154609 / MMMC

1. Installation

Installion follows the instrcuction in fcos https://github.com/tianzhi0549/FCOS/blob/master/INSTALL.md

2.Training and inference

The following command line will train mmmc_R_50_FPN_1x on 8 GPUs with Synchronous Stochastic Gradient Descent (SGD):

python -m torch.distributed.launch \
    --nproc_per_node=8 \
    --master_port=$((RANDOM + 10000)) \
    tools/train_net.py \
    --config-file configs/fcos/mmmc_R_50_FPN_1x.yaml \
    DATALOADER.NUM_WORKERS 2 \
    OUTPUT_DIR training_dir/mmmc_R_50_FPN_1x

The inference command line on coco minival split:

python tools/test_net.py \
    --config-file configs/mmmc/mmmc_R_50_FPN_1x.yaml \
    MODEL.WEIGHT mmmc_R_50_FPN_1x.pth \
    TEST.IMS_PER_BATCH 4    

3. Main result

The main result evaluated on coco test_dev with 2x training scheme.

backbone	ms_train	AP	AP50	AP75	APS	APM	APL	File
R50	Y	43.0	60.6	46.6	25.1	46.5	54.0	model.pth
R50_dcn	Y	44.9	62.7	48.8	25.9	48.6	57.7	model.pth
R101	Y	46.3	64.3	50.3	27.9	50.2	57.9	model.pth
R101_dcn	Y	48.0	66.2	52.0	28.9	51.6	61.1	model.pth
X_101_64x4d	Y	48.5	67.1	52.6	30.7	52.2	60.2	model.pth
X_101_64x4d_dcn	Y	49.9	68.6	54.2	31.2	53.6	62.8	model.pth