yxlijun/s3fd.pytorch

SFD implement with pytorch

s3fd.pytorch's Issues

wider face ap

Hi, glad to see your update.
I saw this issue a few days ago, did you figure out what way the key?
sfzhang15/SFD#57

Testing time is 0.5 seconds per frame using GeForce RTX 3080

Hi, thank you very much for your open source code.
However, I run the 'demo.py' and find the running on one frame using RTX 3080 is about 0.5 seconds. The original paper claimed this method is real-time.

Could you please give me some advice on this?

have you try to change the backbone just like resnet?

Downloading checkpoints

I am finding it real hard to download checkpoint files from the website you uploaded in as I do not understand Chinese.
Is there any other way you can share them? Much appreciated. Thanks.

加载vgg16预训练模型权重报错

作者你好，我在运行train.py时程序在加载vgg16预训练模型权重时报错

我已经从pytorch官方的github地址下载了vgg16的预训练模型，将其放置在指定的weights文件下

现在不知道如何下手，希望您能指导一下，谢谢您了

how to evaluate meanAP of wider face?

Hi,
Thanks for providing the code and pretrained model. However, when I ran the wider_test.py, the meanAP was not shown. And I only found eval_hand.py and eval_head.py in tools/.

Would you please offer the evaluation file of wider face so that we can compare the results with the paper?

What's the frame rate?

anchor_matching_test.py 中 的S3FDValTransform有么？

anchor_matching_test.py 中 的S3FDValTransform有么？

Question about data processing and loss

Hi,
Thanks for sharing the code!
Here are my questions:

1. It seems you use data anchor sampling in your augmentation. Did you test your model without it?
2. What are your calssification loss value and localization loss value in the end of training?

Did you try to replace VGG16 using Res50?

您好,您有尝试将VGG16替换为Res50吗?我之前复现了SFD(F)部分,,test时没有用多尺度,hard子集可以达到79.3%.但替换Res50后,准确率反而下降到了73.6%.想请教一下您这是啥原因.Res50直接用的trochvision里的预训练模型.
另外您的QQ需要真实姓名验证才能加.

Head detection model error!

I have managed to run your face model. But there is a failure of your head model.

would you like to help me out?

运行demo.py时报错

已经下载了readme提供的人脸模型，放置在根目录下的weights目录下

作者可以帮我看下哪里出错了么，谢谢了

How to solve these problems in MATLAB when I plot the curves ?

I ran wider_test.py and got a folder —— s3fd_val in eval_tools.
Then I ran wider_eval.m in MATLAB, but some errors were reported.

---

错误使用 legend (line 273)
未知的属性 'show'

出错 plot_pr (line 22)
legend1 = legend(lendge_name,'show');

出错 wider_plot (line 33)
plot_pr(propose, recall, name_list, seting_class, set_list{i},dateset_class);

出错 wider_eval (line 35)
wider_plot(setting_name_list,dir_int,seting_class,dateset_class);

---

Looking for pretrained weight s3fd.pth in demo.py and base weight vgg16_reducedfc.pth in train.py

Hi, Could you please provide the pretrained weight file for demo.py, so that I can test it on any image? Also the basenet vgg16_reducedfc.pth for training?
Thanks.

can we run this on raspberry pi and jetson nano kind of devices

is there any way to run this on raspberry pi device
what would be the approx fps on it

loss突然变成nan

How to plot the AFW, PASCAL, and FDDB result as shown in the readme file?

License?

What is the license for this code? Could you please clarify so that we decide whether or not to use this code?

Is there mobile version?

This model takes 500+ GFLOPS, it is very hard to land in industry world.

如果要改变某一分支的anchor尺寸，比如说同时使用16/32/64的，类似FaceBoxes，应该如何进行修改？

...由于之前没有做过目标检测，望解答...

Get a model not from pan.baidu.com

It's a problem to download something from pan.baidu.com. May I ask you for uploading pretrained face model on some open resource?

About face detection

How to judge that there is no face?

What should be modified in data/config.py? Can't find file "wider_face_train_bbx_gt.txt"

HI. I am trying to run the code.
Which parts in config.py should be modified?
Error when running prepare_wider_data.py, can not find text file "wider_face_train_bbx_gt.txt".
What is this file and where can I obtain it.
Thank you.

egohands

Hi, is it compatible with your code, do you have any advice on this? Thanks!

a bug in nms

def nms(boxes, scores, overlap=0.5, top_k=200):
keep = scores.new(scores.size(0)).zero_().long()
if boxes.numel() == 0:
return keep
x1 = boxes[:, 0]
y1 = boxes[:, 1]
x2 = boxes[:, 2]
y2 = boxes[:, 3]
area = torch.mul(x2 - x1, y2 - y1)
v, idx = scores.sort(0) # sort in ascending order
# I = I[v >= 0.01]
idx = idx[-top_k:] # indices of the top-k largest vals
xx1 = boxes.new()
yy1 = boxes.new()
xx2 = boxes.new()
yy2 = boxes.new()
w = boxes.new()
h = boxes.new()

# keep = torch.Tensor()
count = 0
while idx.numel() > 0:
    i = idx[-1]  # index of current largest val
    # keep.append(i)
    keep[count] = i
    count += 1
    if idx.size(0) == 1:
        break
    idx = idx[:-1]  # remove kept element from view
    # load bboxes of next highest vals
    torch.index_select(x1, 0, idx, out=xx1)
    torch.index_select(y1, 0, idx, out=yy1)
    torch.index_select(x2, 0, idx, out=xx2)
    torch.index_select(y2, 0, idx, out=yy2)
    # store element-wise max with next highest score
    xx1 = torch.clamp(xx1, min=x1[i])
    yy1 = torch.clamp(yy1, min=y1[i])
    xx2 = torch.clamp(xx2, max=x2[i])
    yy2 = torch.clamp(yy2, max=y2[i])
    w.resize_as_(xx2)
    h.resize_as_(yy2)
    w = xx2 - xx1
    h = yy2 - yy1
    # check sizes of xx1 and xx2.. after each iteration
    w = torch.clamp(w, min=0.0)
    h = torch.clamp(h, min=0.0)
    inter = w * h
    # IoU = i / (area(a) + area(b) - i)
    rem_areas = torch.index_select(area, 0, idx)  # load remaining areas)
    union = (rem_areas - inter) + area[i]
    IoU = inter / union  # store result in iou
    # keep only elements with an IoU <= overlap
    idx = idx[IoU.le(overlap)]
**_return keep, count_**

the two return are not same ,it may take error:
layers/functions/detection.py", line 67, in forward
ValueError: not enough values to unpack (expected 2, got 0)

Consumption of memory is increasing while running the code

Whenever i am passing the images to the model,the memory consumption of RAM is increasing heavily. So is there any way to reduce it? or else this model itself need more configurations?

Thanks in advance.

the inference speed on GPU is slowly

when I download the model you provided ,and test on GPU ,the inference time per_image is about 0.75s ,the picture size is 640*480

weights

Is there a pretrained weight file ?

如果同一张1080p的图用你的模型和论文作者的caffe模型都直接缩放到640*640，测试结果有差别吗，我训练过sfd，但是如果改变了原图的长宽比检测效果就很差，作者的caffe模型就不会，很奇怪啊，按道理数据增强部分都是不改变训练图片的长宽比的呀

Can you give me your contact information?

Hello, I'm a student of UESTC,and I saw your code these days ,and I want to ask you some questions and learn from you

multibox loss

Thanks for your code
in the paper

• the cls loss is normalized by the number of positive and negative anchors
• the reg term is normalized by the number of positive anchors.
  so may be
  
  it should be changed

M = num_pos.data.sum().float() + num_neg.data.sum().float()
loss_c /= N

also in the paper L =\lambda * L_{cls} + L_{reg}
in the code

loss_l, loss_c = criterion(out, targets)
loss = loss_l + loss_c

may be the two place should be modified

Questions about facemodel

Hi, I have downloaded your ''face model'' which is provided in README, and successfully run it on my machine. But now I still have some questions because it seems works well on faces but not work well on heads.
Here is my question: Is the "face model" trained with both head dataset "SCUT-HEAD" and face dataset "widerface", or just only trained with face dataset "widerface"?

The smallest faces in Wider face

您好,向您请教一个额外的问题.
wider face那篇论文中说忽视10像素以下的人脸,但实际标注的人脸有很多低于10像素的,甚至有2个像素的.HR那篇论文中也说人脸像素大于10.
请问是wider face数据集是以前没有标注低于10像素的么?还是说训练时忽视掉低于10像素的人脸?