How to reproduce the 48.5 mAP using the provided scripts test_ap_on_coco.py

[ZYJIQVV]

I have run the command below and got a much low result:

Could you please tell me what the problem is and how to reproduce the result reported in the paper.

[ZYJIQVV]

It seems that batch size causes a significant impact on the results. When I set batch size to 1, the results are perfectly reproduced. But it is so unacceptably slow!

[LiAngCheng-nwpu]

批次大小似乎对结果有重大影响。当我将批量大小设置为 1 时，结果会完美再现。但它的速度太慢了，令人无法接受！

hello,dear can you provide your detailed config for the prefect reproduced?That is very import for me, i am looking forward for your reply ,thank you!

[ZYJIQVV]

批次大小似乎对结果有重大影响。当我将批量大小设置为 1 时，结果会完美再现。但它的速度太慢了，令人无法接受！

hello,dear can you provide your detailed config for the prefect reproduced?That is very import for me, i am looking forward for your reply ,thank you!

It has been such a long time that I can't remember clearly what code I used. But it might this:

CUDA_VISIBLE_DEVICES=3 \
python demo/test_ap_on_coco.py \
 --anno_path /.../coco/annotations/instances_val2017.json \
 --image_dir /.../coco/val2017 \
 --config_file /.../groundingdino/config/GroundingDINO_SwinT_OGC.py \
 --checkpoint_path /.../groundingdino/GroundingDINO/weights/groundingdino_swint_ogc.pth \
 --batch_size 1 \
 --device cuda

And you said config, is it the GroundingDINO_SwinT_OGC.py file?:

batch_size = 1
modelname = "groundingdino"
backbone = "swin_T_224_1k"
position_embedding = "sine"
pe_temperatureH = 20
pe_temperatureW = 20
return_interm_indices = [1, 2, 3]
backbone_freeze_keywords = None
enc_layers = 6
dec_layers = 6
pre_norm = False
dim_feedforward = 2048
hidden_dim = 256
dropout = 0.0
nheads = 8
num_queries = 900
query_dim = 4
num_patterns = 0
num_feature_levels = 4
enc_n_points = 4
dec_n_points = 4
two_stage_type = "standard"
two_stage_bbox_embed_share = False
two_stage_class_embed_share = False
transformer_activation = "relu"
dec_pred_bbox_embed_share = True
dn_box_noise_scale = 1.0
dn_label_noise_ratio = 0.5
dn_label_coef = 1.0
dn_bbox_coef = 1.0
embed_init_tgt = True
dn_labelbook_size = 2000
max_text_len = 256
text_encoder_type = "bert-base-uncased"
use_text_enhancer = True
use_fusion_layer = True
use_checkpoint = True
use_transformer_ckpt = True
use_text_cross_attention = True
text_dropout = 0.0
fusion_dropout = 0.0
fusion_droppath = 0.1
sub_sentence_present = True