Comments (10)
codeAC29
commented on August 18, 2024
@Timo-hab add --dev cuda in command line if you want to run demo on cuda.
from enet-training.
Timo-hab
commented on August 18, 2024
@codeAC29 Thank you for your answer. Unfortunately it does not change anything. In opts.lua it also says cuda is default. I also can see that it takes some GPU memory (~0.8 GB), when I start it.
Have you another idea what the reason could be?
from enet-training.
codeAC29
commented on August 18, 2024
@Timo-hab you will not not get ~150fps while visualization. Visualization has mainly following parts:
- Loading frame
- Forward pass of each frame
- Calculating the winners
- Overlaying output on each frame for displaying
- Displaying the output
In the paper we talk only about processing part (bullet 2) because rest of the steps do not depend on model performance. You can use -v option to see those values in details.
from enet-training.
Timo-hab
commented on August 18, 2024
I add -v and get the following output:
Read : 8 Process : 13 Winner : 20 Colormap: 65 Display : 10 [fps] : 8,61
for a 1000x440px image. The processing time is 13ms which is really really fast.
Interestingly, the process time increases when I reduce the image resolution to half (500x220px):
Read : 4 Process : 20 Winner : 3 Colormap: 16 Display : 3 [fps] : 21,34
How can that be?
from enet-training.
codeAC29
commented on August 18, 2024
I think you are making some mistake somewhere because in my case processing time does reduce when I reduce the image resolution. Try running the demo on any video (eg. of resolution 1000x440) with option -r 1 and run it and then next time use -r 0.5 for same video.
from enet-training.
Timo-hab
commented on August 18, 2024
@codeAC29
Actually thats exactly what I did.
qlua demo.lua -i /home/timo/SegNet/Farbvideo.avi -d /home/timo/ENet-training/model/ -r 1 -verbose
leads to following output:
Read : 35 Process : 315 Winner : 5 Colormap: 57 Display : 14 [fps] : 2,35
Read : 4 Process : 13 Winner : 19 Colormap: 57 Display : 10 [fps] : 9,65
Read : 4 Process : 12 Winner : 20 Colormap: 56 Display : 17 [fps] : 9,16
Read : 8 Process : 18 Winner : 15 Colormap: 81 Display : 11 [fps] : 7,47
Read : 5 Process : 13 Winner : 19 Colormap: 59 Display : 10 [fps] : 9,42
Read : 4 Process : 14 Winner : 19 Colormap: 57 Display : 10 [fps] : 9,59
Read : 4 Process : 12 Winner : 18 Colormap: 57 Display : 10 [fps] : 9,85
Read : 4 Process : 13 Winner : 18 Colormap: 56 Display : 11 [fps] : 9,88
Read : 4 Process : 13 Winner : 17 Colormap: 57 Display : 10 [fps] : 9,78
Read : 4 Process : 14 Winner : 18 Colormap: 56 Display : 10 [fps] : 9,75
Read : 4 Process : 13 Winner : 19 Colormap: 59 Display : 11 [fps] : 9,47
Read : 4 Process : 13 Winner : 20 Colormap: 57 Display : 11 [fps] : 9,54
Read : 4 Process : 13 Winner : 18 Colormap: 56 Display : 10 [fps] : 9,94
Read : 4 Process : 12 Winner : 19 Colormap: 55 Display : 10 [fps] : 10,06
Read : 4 Process : 13 Winner : 19 Colormap: 55 Display : 10 [fps] : 9,82
Read : 4 Process : 12 Winner : 19 Colormap: 55 Display : 10 [fps] : 10,02
Read : 4 Process : 12 Winner : 20 Colormap: 56 Display : 10 [fps] : 9,90
Read : 4 Process : 12 Winner : 18 Colormap: 55 Display : 10 [fps] : 10,10
Read : 4 Process : 14 Winner : 20 Colormap: 56 Display : 10 [fps] : 9,64
Read : 4 Process : 13 Winner : 17 Colormap: 55 Display : 9 [fps] : 10,21
Read : 4 Process : 14 Winner : 19 Colormap: 56 Display : 13 [fps] : 9,35
Read : 4 Process : 13 Winner : 20 Colormap: 57 Display : 14 [fps] : 9,30
If I change -r 1 to -r 0.5:
qlua demo.lua -i /home/timo/SegNet/Farbvideo.avi -d /home/timo/ENet-training/model/ -r 0.5 -verbose
leads to following output:
Read : 37 Process : 320 Winner : 2 Colormap: 15 Display : 11 [fps] : 2,60
Read : 9 Process : 36 Winner : 2 Colormap: 25 Display : 4 [fps] : 13,23
Read : 6 Process : 25 Winner : 2 Colormap: 17 Display : 5 [fps] : 18,77
Read : 4 Process : 21 Winner : 2 Colormap: 14 Display : 3 [fps] : 22,37
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 5 [fps] : 22,51
Read : 8 Process : 32 Winner : 2 Colormap: 24 Display : 6 [fps] : 14,10
Read : 6 Process : 24 Winner : 3 Colormap: 16 Display : 6 [fps] : 18,55
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 6 [fps] : 21,53
Read : 4 Process : 21 Winner : 2 Colormap: 15 Display : 5 [fps] : 21,71
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 6 [fps] : 21,40
Read : 4 Process : 20 Winner : 3 Colormap: 14 Display : 7 [fps] : 20,30
Read : 4 Process : 20 Winner : 2 Colormap: 15 Display : 7 [fps] : 21,05
Read : 4 Process : 21 Winner : 3 Colormap: 15 Display : 4 [fps] : 21,32
Read : 10 Process : 37 Winner : 2 Colormap: 26 Display : 4 [fps] : 12,66
Read : 7 Process : 31 Winner : 2 Colormap: 20 Display : 3 [fps] : 16,06
Read : 5 Process : 25 Winner : 2 Colormap: 17 Display : 3 [fps] : 19,43
Read : 5 Process : 23 Winner : 2 Colormap: 15 Display : 3 [fps] : 21,40
Read : 4 Process : 21 Winner : 2 Colormap: 15 Display : 3 [fps] : 22,51
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 5 [fps] : 22,02
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 3 [fps] : 23,05
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 3 [fps] : 22,89
Read : 4 Process : 20 Winner : 3 Colormap: 16 Display : 3 [fps] : 21,84
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 4 [fps] : 22,21
Read : 4 Process : 20 Winner : 3 Colormap: 14 Display : 4 [fps] : 21,61
Read : 4 Process : 21 Winner : 1 Colormap: 14 Display : 3 [fps] : 22,96
Read : 4 Process : 21 Winner : 3 Colormap: 14 Display : 5 [fps] : 21,49
Read : 4 Process : 21 Winner : 2 Colormap: 14 Display : 5 [fps] : 21,33
Read : 4 Process : 21 Winner : 2 Colormap: 14 Display : 6 [fps] : 21,31
Read : 4 Process : 21 Winner : 2 Colormap: 14 Display : 6 [fps] : 21,34
Read : 4 Process : 20 Winner : 3 Colormap: 14 Display : 3 [fps] : 22,54
Read : 4 Process : 20 Winner : 2 Colormap: 14 Display : 3 [fps] : 22,94
Read : 4 Process : 20 Winner : 3 Colormap: 14 Display : 3 [fps] : 22,92
Read : 4 Process : 20 Winner : 3 Colormap: 14 Display : 3 [fps] : 22,94
The video resolution is 1000x440px. I dont understand why it takes for the half resolution nearly twice as much processing time.
from enet-training.
lbin
commented on August 18, 2024
I also got this problem, @Timo-hab do you find the reasons?
from enet-training.
mathTaoTao
commented on August 18, 2024
I have the similar problem on Jetson TX2.My test set is CamVid
qlua demo.lua -i camvid/test/ -d weight/
the forward time is about 37ms using the original resolution(480*360), which is lower compared with your result on TX1 because TX2 is at least twice as fast as TX1.
when I change the resolution,
qlua demo.lua -i camvid/test/ -d weight/ -r 2
the forward time remain almost the same,I don't know what's wrong with it.
from enet-training.
codeAC29
commented on August 18, 2024
@Timo-hab @lbin @xiupingmath I had included rescaling time as well as time taken to move tensor from cpu to gpu in processing time. Now they are separated from forward pass time. f762144
from enet-training.
lbin
commented on August 18, 2024
@codeAC29 Thanks
from enet-training.
Related Issues (20)
- What is the learning rate decay and preprocessing you used in your training? HOT 13
- Consider hosting the pretrained model on Github HOT 1
- model-cityscapes.net released HOT 1
- what is the test time per image on cpu HOT 1
- implementation different from paper? HOT 1
- Do you freeze weights of encoder while training the whole model HOT 1
- encoder weights HOT 5
- "assertion failed!" When trying to run demo.lua
- In frameimage.lua:17: module 'fastimage' not found: HOT 2
- Not able to reproduce the fps on tx2 HOT 3
- Confused about the camVid dataset used to train encoder HOT 5
- Is the error the same as the loss HOT 1
- lua/5.2/nn/JoinTable.lua:38: bad argument #1 to 'copy' (sizes do not match at /home/user/torch/extra/cutorch/lib/THC/THCTensorCopy.cu:31) HOT 1
- add one more class to pre-trained model
- Which encoder weights should I use as CNNEncoder?
- Cityscapes Test Result HOT 1
- Assertion `t >= 0 && t < n_classes` failed, HOT 1
- Training ENet using own Dataset
- there is no mask when i try the visualization
- input and target should be of same size
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from enet-training.