hu64 / spotnet Goto Github PK

I found that your code is no different from centernet. There is more knowledge in the semi-supervised-segmentation folder, and the up-sampling segmentation sub-network is not built as in the paper, and there is almost no explanation on how to call in the readme. If you The code completely complies with the calling rules of centernet, then it doesn’t matter, but can you explain where your improvements are, thank you very much for your help。

Thanks for your great work! Would it be convenient for you to share the code of converting UA DETRAC into COCO format?

Segmentation Head: each block is composed of two
3x3 convolutions followed by upsampling

The added segmentation head takes as input a feature map
that has been reduced by a factor of four in terms of spatial
dimension when compared to the input. It consists of three
3 × 3 convolutions, with upsampling layers in between.

Do these two statments conflit with each other?

We used the stacked hourglass network as our backbone because it shows the best performance for keypoint estimation.

It get high mAP but low FPS, right?

Thx!

We used the stacked hourglass network as our backbone
because it shows the best performance for keypoint estima�tion

Many thx!

I think the attention map make the value of some of the predicted keypoints lower , which means the detection is more strict than CenterNet.
Will the Average Recall become lower?
I do not well understand the difference between coco AP and AR ? Would you mind sharing your explaination?
Many thanks!

To attenuate the response at locations unlikely to contain an object of interest, we multiply every channel of the feature map with our segmentation map, thus reducing the probability of false positives in irrelevant areas.

If the with&height featrue maps are multiplied with the segmentation map, the size will be smaller, right?
thx!

I used the UA-DETRAC dataset to train the CenterNet model, but it did not achieve the performance in your paper. Can you explain your experimental process and parameter settings? thanks!

Sorry to disturb you.
Recently, I am doing some research on the UAV data set, but I found that I cannot find a suitable way to verify the results of my network. The official data does not seem to provide a test set, and it can be verified without submitting it online. Performance online test, so I want to ask how you used the data set in the first place

Please add the steps how to train this model? what is the input and output

Hi, could you please share your files in csv format? i.e. '/store/datasets/UAV/csv.csv' or '/store/datasets/UAV/val.csv' or sharing some code elaborating how the files in csv format are being created? Thanks.

Hey,

I have a question on the difference in APs computed by COCO-API vs computed by the official MATLAB evaluation tool:

• Using COCO-API to compute the AP@IoU=0.7, I get ~78 on test sequences (test_b.json), which is lower than what has been reported using MATLAB evaluation tool.

• I am evaluating your released model: 'ua-detrac_model_best.pth' so no training involved.

• I cannot run MATLAB tool as it is not compatible with my OS.

Thanks :)

Hi, could you please share your annotation files? i.e. '/store/datasets/UA-Detrac/COCO-format/test-1-on-30_b.json' or sharing some code elaborating how the annotations are being created? Thanks.

I trained centerNet on UA-detrac, without considering the ignore region. The trained model can not find small vehicle.
So I think the ignore region is important.

Thx!