IDEA-Research / GroundingDINO

[ECCV 2024] Official implementation of the paper "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection"
https://arxiv.org/abs/2303.05499
Apache License 2.0
5.81k stars 613 forks source link

G-DINO to ONNX #156

Open kobic8 opened 1 year ago

kobic8 commented 1 year ago

is there a way to export the Grounding DINO part to ONNX?

oylz commented 1 year ago

here

kobic8 commented 1 year ago

here

is there a code for that?

wangshuang-jiayou commented 1 year ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

oylz commented 1 year ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

yes.the same as the original model's result.

wangshuang-jiayou commented 1 year ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

yes.the same as the original model's result.

Thanks for your replay!!!I found the reason: I forgot to sigmoid the logits result. But got a little bit low class confidence~Could you provide the code? THANK YOU VERY MUCH~~~

oylz commented 1 year ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

yes.the same as the original model's result.

Thanks for your replay!!!I found the reason: I forgot to sigmoid the logits result. But got a little bit low class confidence~Could you provide the code? THANK YOU VERY MUCH~~~

the sigmoid call is goes like this

float sigmoid = 1.00/(1 + exp(-logits));
wangshuang-jiayou commented 11 months ago

here

tensorrt doesn't support int64. Does the int64 in onnx casted to int32 in tensorrt influence the result in your experiment? My onnx results is right,but tensorrt result is wrong

ganliqiang commented 8 months ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

yes.the same as the original model's result.

Thanks for your replay!!!I found the reason: I forgot to sigmoid the logits result. But got a little bit low class confidence~Could you provide the code? THANK YOU VERY MUCH~~~

do you infer the onnx with tensorrt ?can you show your code? @wangshuang-jiayou

xiyangyang99 commented 7 months ago

here

tensorrt doesn't support int64. Does the int64 in onnx casted to int32 in tensorrt influence the result in your experiment? My onnx results is right,but tensorrt result is wrong

hello big shot,can you share your inference onnx scripts?thinks!

xiyangyang99 commented 6 months ago

here

tensorrt doesn't support int64. Does the int64 in onnx casted to int32 in tensorrt influence the result in your experiment? My onnx results is right,but tensorrt result is wrong

大佬你的tensorrt转换脚本能分享一下吗?

Thivish16 commented 4 months ago

Hi, It would be more helpful if you guys share the conversion logic that incorporated for converting to .onnx. I am now bit confused on the inference with onnx runtime. Thanks for the help!

blacksino commented 4 months ago

here

image Hi~I have infer the onnx you provide,but got negative logits. Is it normoal?

yes.the same as the original model's result.

Thanks for your replay!!!I found the reason: I forgot to sigmoid the logits result. But got a little bit low class confidence~Could you provide the code? THANK YOU VERY MUCH~~~

do you infer the onnx with tensorrt ?can you show your code? @wangshuang-jiayou

你好,请问你解决这个问题了么,我转成onnx发现数值对不上