dlyldxwl / fssd.pytorch Goto Github PK
View Code? Open in Web Editor NEWPytorch re-implementation of Fssd
Pytorch re-implementation of Fssd
when i train VOC2007,it work ,but the progress is terrible
Epoch:1 || epochiter: 0/1034|| Totel iter 0 || L: 201.1674 C: 330.5661 S: 531.7335||Batch time: 12.2342 ||LR: 0.0010000 Epoch:1 || epochiter: 10/1034|| Totel iter 10 || L: nan C: nan S: nan||Batch time: 0.8318 ||LR: 0.0010000 Epoch:1 || epochiter: 20/1034|| Totel iter 20 || L: nan C: nan S: nan||Batch time: 0.8154 ||LR: 0.0010000 Epoch:1 || epochiter: 30/1034|| Totel iter 30 || L: nan C: nan S: nan||Batch time: 0.8430 ||LR: 0.0010000 Epoch:1 || epochiter: 40/1034|| Totel iter 40 || L: nan C: nan S: nan||Batch time: 0.8290 ||LR: 0.0010000 Epoch:1 || epochiter: 50/1034|| Totel iter 50 || L: nan C: nan S: nan||Batch time: 0.8428 ||LR: 0.0010000 Epoch:1 || epochiter: 60/1034|| Totel iter 60 || L: nan C: nan S: nan||Batch time: 0.8450 ||LR: 0.0010000 Epoch:1 || epochiter: 70/1034|| Totel iter 70 || L: nan C: nan S: nan||Batch time: 0.8438 ||LR: 0.0010000 Epoch:1 || epochiter: 80/1034|| Totel iter 80 || L: nan C: nan S: nan||Batch time: 0.8202 ||LR: 0.0010000 Epoch:1 || epochiter: 90/1034|| Totel iter 90 || L: nan C: nan S: nan||Batch time: 0.8445 ||LR: 0.0010000 Epoch:1 || epochiter: 100/1034|| Totel iter 100 || L: nan C: nan S: nan||Batch time: 0.8235 ||LR: 0.0010000 Epoch:1 || epochiter: 110/1034|| Totel iter 110 || L: nan C: nan S: nan||Batch time: 0.8498 ||LR: 0.0010000 Epoch:1 || epochiter: 120/1034|| Totel iter 120 || L: nan C: nan S: nan||Batch time: 0.8209 ||LR: 0.0010000 Epoch:1 || epochiter: 130/1034|| Totel iter 130 || L: nan C: nan S: nan||Batch time: 0.8657 ||LR: 0.0010000 Epoch:1 || epochiter: 140/1034|| Totel iter 140 || L: nan C: nan S: nan||Batch time: 0.8319 ||LR: 0.0010000 Epoch:1 || epochiter: 150/1034|| Totel iter 150 || L: nan C: nan S: nan||Batch time: 0.8366 ||LR: 0.0010000 Epoch:1 || epochiter: 160/1034|| Totel iter 160 || L: nan C: nan S: nan||Batch time: 0.8760 ||LR: 0.0010000 Epoch:1 || epochiter: 170/1034|| Totel iter 170 || L: nan C: nan S: nan||Batch time: 0.8323 ||LR: 0.0010000 Epoch:1 || epochiter: 180/1034|| Totel iter 180 || L: nan C: nan S: nan||Batch time: 0.8389 ||LR: 0.0010000 Epoch:1 || epochiter: 190/1034|| Totel iter 190 || L: nan C: nan S: nan||Batch time: 0.8319 ||LR: 0.0010000 Epoch:1 || epochiter: 200/1034|| Totel iter 200 || L: nan C: nan S: nan||Batch time: 0.8414 ||LR: 0.0010000 Epoch:1 || epochiter: 210/1034|| Totel iter 210 || L: nan C: nan S: nan||Batch time: 0.8519 ||LR: 0.0010000
could you help me?
I notice that for the backbone network vgg_prune, convolutions have BN but no bias, could you tell me where I could find the pre-trained vgg for FSSD_VGG_BN?
Thank you!
thx
im very interested in this project you make,but im a greenhand in object detection,can you tell me how to train?
Dear @dlyldxwl ,
Great work! have you ever tried focal loss in FSSD? It seems to be better than ohem in your programme.
Following repos might help:
https://github.com/marvis/pytorch-yolo2/blob/master/FocalLoss.py
https://github.com/yhenon/pytorch-retinanet/blob/master/losses.py
I have tried myself, but it seems to be worse than ohem...
Best regards,
Traceback (most recent call last):
File "train.py", line 268, in
train()
File "train.py", line 233, in train
loss.backward()
File "/home/huhuai/anaconda3/lib/python3.6/site-packages/torch/tensor.py", line 93, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/home/huhuai/anaconda3/lib/python3.6/site-packages/torch/autograd/init.py", line 90, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
运行报错:local variable 'result' referenced before assignment
result之前没定义呢
Hi...Is the pruning for 300X300 only? I didnt find any pruning for 512X512. Thanks
您好,我有两个问题想请教一下。
第一个是train.py第114行,torch.load(resume_net)这里的resume_net没定义吧
第二个问题是test.py运行时报错RuntimeError: Error(s) in loading state_dict for FSSD:size mismatch for conf.0.bias: copying a param with shape torch.Size([126]) from checkpoint, the shape in current model is torch.Size([486])........
这里net.load_state_dict(new_state_dict)加载fssd_voc_79_74.pth导致了size不匹配
when i find some wrong i can't solve
`
rw@rw:/新建文件夹/fssd.pytorch-master$ source activate pytorch0.4.0/新建文件夹/fssd.pytorch-master$ python demo.py
(pytorch0.4.0) rw@rw:
Finished loading model
Traceback (most recent call last):
File "demo.py", line 114, in
test_net(net, img, img_name, detector, transform, priors,top_k=200, thresh=0.4)
File "demo.py", line 38, in test_net
out = net(x,test=True)
File "/home/rw/anaconda3/envs/pytorch0.4.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/rw/新建文件夹/fssd.pytorch-master/models/FSSD_VGG.py", line 105, in forward
transformed_features.append(v(source_features[k]))
File "/home/rw/anaconda3/envs/pytorch0.4.0/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/home/rw/新建文件夹/fssd.pytorch-master/models/FSSD_VGG.py", line 28, in forward
x=F.interpolate(input=x,size=(self.up_size, self.up_size), mode='bilinear')
AttributeError: module 'torch.nn.functional' has no attribute 'interpolate'
`
i have put weight into the folder
but something is wrong ?
would you help me??
I am sorry to bother you.
I apply the way of pruning for the other object detection network. But the FPS reduced after pruning.
(pytorch0.4.0) rw@rw:~/fssd.pytorch-master$ python demo.py
Traceback (most recent call last):
File "demo.py", line 13, in
from data import VOCroot,COCOroot
File "/home/rw/fssd.pytorch-master/data/init.py", line 3, in
from .coco import COCODetection
File "/home/rw/fssd.pytorch-master/data/coco.py", line 21, in
from utils.pycocotools.coco import COCO
File "/home/rw/fssd.pytorch-master/utils/pycocotools/coco.py", line 55, in
from . import mask as maskUtils
File "/home/rw/fssd.pytorch-master/utils/pycocotools/mask.py", line 4, in
from . import _mask
ImportError: /home/rw/fssd.pytorch-master/utils/pycocotools/_mask.cpython-35m-x86_64-linux-gnu.so: undefined symbol: PyFPE_jbuf
请问大佬,我看了您的csdn,所以慕名前来。但是我想尝试图像结果可视化,不知道如何写,请问大佬能不能提供思路帮助下我,刚上路,感觉还是看着图片可视化来的舒服,望交流回复
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.