Jian comments

Results 39 comments of


                                            Jian

Is it possible to use smaller GPU for inference?

Hi @cqlyiyeshu I think that 16G is enough. you can try to use less scale in demo.py. especially for 2x.

Where is IAM module

IAM module is not a network structure, it is a combination between anchor design and gt face sampling. you can find IAM in data augmentation and anchor designing code.

Can I know which of the options in `widerface_640` config are toggled during training?

you can try different combination of these configures

What is the batch size when you train and with what and how many GPUs?

2GPUs, and each gpu has 8 images.

why First shot PAL is Not Using in this code?

First shot PAL is not included in this inference code.

Can I know if you use the exact same data augmentation as amdegroot/ssd.pytorch?

Besides standard SSD data augmentation, we still use anchor-based sampling strategy.

pretrain model selection?

Hi @mathyyj imagenet resnet pretrained model. This pretrained model can be imported by torchvision.

Problem when finetuning with ResNet caffemodel

can you tell me where can i download the resnet caffemodel for finetuning

ValueError: not enough values to unpack (expected 2, got 0)

原因是检测的结果为空，就是boxes是空的，需要添加一个判断