Questions tagged [pytorch]
Pytorch is an open source library for Tensors and Dynamic neural networks in Python with strong GPU acceleration. For details, see https://pytorch.org.
708 questions
0 votes
0 answers
6 views
Sequence generation model produces incorrect, but coherent outputs
My model takes in an image of a handwritten equation and converts it into its LaTeX representation. In order to do this, it uses a ResNet50 pre-trained model for feature extraction and a transformer ...
- 71
7 votes
1 answer
112 views
LSTM feature scaling with windowing?
Beginner ML practitioner here. I'm trying to do some time series forecasting on a fairly high resolution dataset that stretches over a long period of time. The values vary pretty widely over time: to ...
- 251
2 votes
0 answers
40 views
Fine-tuning YOLO: Directly cloning and modifying the GitHub repo vs. using Transformers library and Hugging Face — pros and cons?
I’m planning to fine-tune a YOLO model for a custom object detection task. There seem to be two main approaches: Clone the official YOLO GitHub repository (e.g., YOLOv5 or YOLOv8), adjust the codebase ...
- 141
3 votes
1 answer
47 views
Single nn.Embedding instance vs mulitple nn.Embedding instances
I am trying to determine if using multiple instances of nn.Embedding() has any value over using a single instance in training a model. As an example, let's say I ...
- 33
0 votes
0 answers
16 views
terrible performance on CIFA10 using SWIN model
I am trying to apply the idea from Embedding Deep Networks into Visual Explanations and see if it works on Transformers. The performance is terrible because the accuracy hasn't passed 10%. Can someone ...
- 101
3 votes
1 answer
54 views
Model seems to peek into target sequence and cheat during training despite using masking
I am using CNN-transformer hybrid architecture to detect handwritten equation and convert them to LaTex strings. All target sequences (the actual LaTex representation of a handwritten equation) are ...
- 71
0 votes
1 answer
57 views
when testing with shuffled data, accuracy is high, but when testing with unshuffled data, accuracy is low
To be clear, I shuffled my data when I trained it. It is only the testing data that I modified to be unshuffled, and found that accuracy tanks. (i also used the same data for training and for testing)
- 71
0 votes
1 answer
37 views
Why is my upscaling gan not working?
I have been trying to code an upscaling gan but while the code run, I pretty much always end up with terrible result when the gan doesn't collapse, collapse which happen often. I previously tried to ...
- 1