-
My code to run Deep Explainer on some Imagenet images is as follows:
```python3
import glob
# load the model
model = models.resnet50(pretrained=True).eval()
# Get the prediction for the input i…
-
I try to compress mixnet.py, Can you check again case concat-split, I try to unpruned conv before concat, and after split. But they return false in split. Thank's you
[MixNet_zig.gv.pdf](https://gith…
-
UPDATE: We only miss the conv1d interface.
We need a new interface. I'll give more detail later today.
Include a conv1d in the interface.
Make the make_thunk print the stack trace that create…
nouiz updated
7 years ago
-
In your `conv2d_keep_shape` function which I copied below what is the purpose of the `image_shape` and `filter_shape` args?
```
def conv2D_keep_shape(x, w, image_shape, filter_shape, subsample=(1, 1…
-
At this moment, we only support `Dense` layers. This is becuase `SparseArrays.jl` currently supports only 1D vectors or 2D matrices, so we could try transforming `Conv` layers into dense matrix multip…
-
### Motivation and description
Is there such a thing as too many activation functions? I haven't decided on which of these, or if all:
**Natural Logarithm rescaled ReLU (NLReLU)**
https://arxiv…
-
Hi, I want to ask a simple problem about the mnist classification example. Images in mnist are treated as sequences by expanding them to 1D, then each image gets a probability distribution. But imag…
-
hello, in **quant_train_module.py** file, i saw a line of code : y.data.copy_(yq.data), this code change the data of relu's output data.data, in order to use it in backword for calculate activation's…
-
It appears that the batchNormWeight and batchNormBias are being updated to be equal to what I'd expect the runningVar and runningMean to be.
```
BatchNorm {batchNormWeight = IndependentTensor {toD…
-
##### System information (version)
- OpenCV => 4.2.0
##### Detailed description
The CUDA backend can support mixed-precision inference with various types: FP32, FP16, INT32, (U)INT8 and possibl…