Official repository for the paper TRIDENT: Transductive Decoupled Variational Inference for Few Shot Classification
Home Page: https://openreview.net/forum?id=bomdTc9HyL
License: MIT License
Hi @anujinho,
I am trying to reproduce your TRIDENT CCVAE model, however, I am not able to pass inputs through the learner/model.
Below is the model:
MAML(
(module): CCVAE(
(encoder): CEncoder(
(net): Sequential(
(0): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): LeakyReLU(negative_slope=0.2)
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): LeakyReLU(negative_slope=0.2)
(7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): LeakyReLU(negative_slope=0.2)
(11): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(12): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(14): LeakyReLU(negative_slope=0.2)
(15): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(16): Flatten(start_dim=1, end_dim=-1)
)
)
(decoder): CDecoder(
(linear): Sequential(
(0): Linear(in_features=128, out_features=800, bias=True)
(1): LeakyReLU(negative_slope=0.2)
)
(net): Sequential(
(0): UpsamplingNearest2d(size=(10, 10), mode=nearest)
(1): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=same)
(2): LeakyReLU(negative_slope=0.2)
(3): UpsamplingNearest2d(size=(21, 21), mode=nearest)
(4): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=same)
(5): LeakyReLU(negative_slope=0.2)
(6): UpsamplingNearest2d(size=(42, 42), mode=nearest)
(7): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=same)
(8): LeakyReLU(negative_slope=0.2)
(9): UpsamplingNearest2d(size=(84, 84), mode=nearest)
(10): Conv2d(32, 3, kernel_size=(3, 3), stride=(1, 1), padding=same)
(11): Sigmoid()
)
)
(classifier_vae): Classifier_VAE(
(encoder): TADCEncoder(
(net): Sequential(
(0): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): LeakyReLU(negative_slope=0.2)
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): LeakyReLU(negative_slope=0.2)
(7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): LeakyReLU(negative_slope=0.2)
(11): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(12): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(14): LeakyReLU(negative_slope=0.2)
(15): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(fe): Sequential(
(0): Conv2d(1, 64, kernel_size=(110, 1), stride=(1, 1), padding=valid, bias=False)
(1): LeakyReLU(negative_slope=0.2)
(2): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1), padding=valid, bias=False)
(3): LeakyReLU(negative_slope=0.2)
)
(f_q): Conv2d(32, 1, kernel_size=(1, 1), stride=(1, 1), padding=valid, bias=False)
(f_k): Conv2d(32, 1, kernel_size=(1, 1), stride=(1, 1), padding=valid, bias=False)
(f_v): Conv2d(32, 1, kernel_size=(1, 1), stride=(1, 1), padding=valid, bias=False)
)
(gaussian_parametrizer): GaussianParametrizer(
(h1): Linear(in_features=864, out_features=64, bias=True)
(h2): Linear(in_features=864, out_features=64, bias=True)
)
(classifier): Sequential(
(0): Linear(in_features=64, out_features=32, bias=True)
(1): LeakyReLU(negative_slope=0.2)
(2): Linear(in_features=32, out_features=10, bias=True)
)
)
(gaussian_parametrizer): GaussianParametrizer(
(h1): Linear(in_features=800, out_features=64, bias=True)
(h2): Linear(in_features=800, out_features=64, bias=True)
)
)
)
And below is the minimized stack trace:
Traceback (most recent call last):
...
File ".../trident.py", line 105, in _train_epoch
eval_loss, eval_acc = inner_adapt_trident(ttask, reconst_loss,
File ".../utils.py", line 146, in inner_adapt_trident
reconst_image, logits, mu_l, log_var_l, mu_s, log_var_s = learner(
...
File ".../archs.py", line 812, in forward
mu_s, log_var_s = self.gaussian_parametrizer(xs)
...
File ".../archs.py", line 404, in forward
mu = self.h1(x)
...
RuntimeError: mat1 and mat2 shapes cannot be multiplied (10x128 and 800x64)
Also, below are the hyperparameters:
n_ways: 5
k_shots: 1
q_shots: 10
meta_batch_size: 20
order: False
inner_lr: 0.0014
task_adapt: True
zl: 64
zs: 64
reconstr: std
dataset: cifarfs
wm_channels: 64
wn_channels: 32
download: False
extra: False
adapt_steps_train: 5
adapt_steps_test: 5
Basically the output shape of CEncoder and input shape of GaussianParametrizer doesn't match.
Can you please help to resolve this issue. Will look forward to hearing from you soon. Thanks!
Dear Anuj,
I am very impressed from your work and I would like to try and replicate your training to understand how your method works.
I have completed the training and testing of my own data, but when I want to reason and test an image, there is a problem. I looked at the algorithm and found that each reason needs to input multiple images, multiple prediction sets and query sets image, how can I realize the inference of a single image and obtain the prediction result of a single image?
Dear Anuj,
I am very impressed from your work and I would like to try and replicate your training to understand how your method works. I was wondering how did you manage to train the models because it seems that it will take me 6-7 days to train a model using the iterations set-up on configs. did you parallelize it somehow and if yes do you have any instructions for that?
Regards,
Michalis
size mismatch for module.classifier_vae.encoder.fe.0.weight: copying a param with shape torch.Size([64, 1, 30, 1]) from checkpoint, the shape in current model is torch.Size([64, 1, 10, 1]).
React
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
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
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.
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.