CenterNet-Lite icon indicating copy to clipboard operation
CenterNet-Lite copied to clipboard

A resnet18 version of CenterNet(objects as points)

Results 9 CenterNet-Lite issues
Sort by recently updated
recently updated
newest added

以前的代码能上70 现在最新的代码死活都只有40几的MAP是怎么回事呢

在使用torch.nn.DataParallel多gpu训练的时候,train阶段没问题,val阶段报错。 raceback (most recent call last): File "train.py", line 347, in train() File "train.py", line 315, in train evaluator.evaluate(model) File "/home/11/CenterNet-Lite-master/utils/cocoapi_evaluator.py", line 196, in evaluate outputs = model(x) File "/home/11/.local/lib/python3.8/site-packages/torch/nn/modules/module.py",...

仿照cocodataset写自定义数据集加载,在自定义数据集的时候,比如类别有0和1,class_labels和class_index是否需要加入背景这一类?

![det1](https://user-images.githubusercontent.com/34679219/105323094-b159a980-5c04-11eb-985c-e4d011449f42.png)

topk_scores = topk_scores[0].cpu().numpy() topk_ind = topk_clses[0].cpu().numpy() topk_bbox_pred = bbox_pred[topk_inds[0]].cpu().numpy() 为什么要去取[0]啊,我不太理解

有没有比较过使用spp提高了多少? spp能提高的来源呢? 再就是没有使用多尺度融合吗?