bit-da / o2net Goto Github PK

The calculation method explained in the paper is to assign a weight of 1 to the position within the bounding boxes in the mask matrix, and assign a weight of 0 to the position outside the bounding boxes. But in your code, 1 is added to the mask matrix first, and then divided by the mean. Such a matrix is not a matrix containing only 0 and 1.

I'm very interested in this work and have been attempting to reproduce the results. However, after following the instructions in the README, I have been unable to reproduce a mAP score of 47.2 after 50 epochs of training. Specifically, after pretraining a source model, then running domain adaptive training using the pretrained source model, I get a mAP score of 39.69. I trained using 1 GPU and a batch size of 2, the result of the experiment are shown below.

However, I do not have the exact same system setup as described in the README. I'm using newer versions of Pytorch, Torchvision, and CUDA because my graphics card is too new and is not compatible with CUDA 9.2 anymore.

If someone could point me in the right direction to reproduce the results of this paper, that would be great!

IoU metric: bbox
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.21836
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.39669
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.20980
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.04972
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.22982
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.40898
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.18241
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.33332
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.35851
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.09898
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.35826
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.61922

look forward to your reply.

你好，实验是需要下载预训练权重吗，在哪里下载哪个？我的自定义数据集里AP一直为0

Hi , I recently cloned the O2net repo and setup the env, but while trying to build the model I am facing an issue with "ModuleNotFoundError: No module named 'MultiScaleDeformableAttention'" .
Can you please help ho to bypass this issue ?

Only provide cityscapes dataloader?

The ops directory is created outside of models directory in this repo, needs correction as in original deformable-detr the ops is inside models

Great work I am also trying to replicate the paper!