Comments (7)
Try changing line 33 in cudamat/cudamat_conv_util.cuh
#define TEXTURE_SIZE_MAX (1<<29)
to
#define TEXTURE_SIZE_MAX 0
from convnet.
Thanks Nitish!
Unfortunately it didn't work because the call to getTextureObject
happens from convContrastNormCrossMap
in cudamat_conv_others.cu
:
if (blocked) {
if (checkCaseBounds) {
cudaFuncSetCacheConfig(kFCNorm<4, 32, 4, true, true>, cudaFuncCachePreferL1);
kFCNorm<4, 32, 4, true, true><<<blocks, threads, 0, stream>>>(getTextureObject(images), getTextureObject(meanDiffs), target->data_device,
imgSize, numFilters, numImages, sizeF, addScale, powScale, minDiv);
} else {
cudaFuncSetCacheConfig(kFCNorm<4, 32, 4, false, true>, cudaFuncCachePreferL1);
kFCNorm<4, 32, 4, false, true><<<blocks, threads, 0, stream>>>(getTextureObject(images), getTextureObject(meanDiffs), target->data_device,
imgSize, numFilters, numImages, sizeF, addScale, powScale, minDiv);
}
} else {
if (checkCaseBounds) {
//!!!!!!!!!!!!!!!!!! HERE !!!!!!!!!!!!!!!!!!!!!!!!!!!!
cudaFuncSetCacheConfig(kFCNorm<4, 32, 4, true, false>, cudaFuncCachePreferL1);
kFCNorm<4, 32, 4, true, false><<<blocks, threads, 0, stream>>>(getTextureObject(images), getTextureObject(meanDiffs), target->data_device,
imgSize, numFilters, numImages, sizeF, addScale, powScale, minDiv);
} else {
cudaFuncSetCacheConfig(kFCNorm<4, 32, 4, false, false>, cudaFuncCachePreferL1);
kFCNorm<4, 32, 4, false, false><<<blocks, threads, 0, stream>>>(getTextureObject(images), getTextureObject(meanDiffs), target->data_device,
imgSize, numFilters, numImages, sizeF, addScale, powScale, minDiv);
}
}
What would be the easiest way to fix this?
from convnet.
Looks like it needs a kernel that doesn't use textures. I will write one and push it up soon.
from convnet.
to Nitish,
I also met the problem.
And now I regard it as a suspicious thing that,
the datatype of MNIST input(:*.h5) is 32bit floating and the range is [0, 1]
but for imagenet case (;be converted via jpeg2hdf5)
the one is 8 bit unsigned integer and the range is(;be looks) [0, 255].
could you review the thing?
from convnet.
@regata @neuralix
The latest commit adds a new kernel that doesn't use textures.
from convnet.
@neuralix The datatypes are fine. Everything gets converted to floats when it is moved to the GPU so it doesn't matter. The images are normalized and mean-centered later.
from convnet.
@nitishsrivastava I've met same problem at same point.
Now I doubt two points. The one is difference btn. my GPU(:Kepler) and your (recommended) one:Fermi. And the other is the issue of range in training data mentioned before. (I'll report the result.)
from convnet.
Related Issues (20)
- How to cite it? HOT 1
- meaning of variables in RESPONSE_NORM type edge HOT 1
- imagenet case
- Assertion error about targets.shape[3] during Backprop. HOT 1
- about 'partial_sum'
- Warning : Could not set up P2P, GPU-to-GPU communication will be slow. CUDA error
- Step 1Invalid start / end 0 0 HOT 2
- video2hdf5.o error HOT 2
- Comipling the CPU-only feature extractor HOT 3
- Bug in LoadChunk(DataIterator& it, Matrix& mat, vector<int>& random_rows) HOT 1
- Unable to compile: undefined references HOT 1
- attempting to build cpu-only on OSX 10.10 HOT 1
- Wrong post
- GetName method undefined in pbtxt2dot.py
- Windows support
- Can we get some walk through on how to train CNN? HOT 1
- Assertion failed when run py/run_convnet.py
- Display localization
- How to train the convnet on multiGPUs?
- /usr/bin/ld: cannot find -leigenmat HOT 1
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 convnet.