From cb7315b1661708f7ee2e77d61fc3fa2c8f1ec7d0 Mon Sep 17 00:00:00 2001 From: David Date: Wed, 8 May 2019 16:13:40 +0200 Subject: [PATCH] restrucured in the readmes --- README.md | 36 +++++------------------------------- datasets/README.md | 15 +++++++++++++++ inspiration/README.md | 8 ++++++++ 3 files changed, 28 insertions(+), 31 deletions(-) create mode 100644 datasets/README.md create mode 100644 inspiration/README.md diff --git a/README.md b/README.md index ca2f182..2a28b74 100644 --- a/README.md +++ b/README.md @@ -4,19 +4,16 @@ Make sure you have `virtualenv` and `python 3.5+` installed - ```bash bash install.sh ``` This will activate the virtualenv and install the proper packages. -You will probably also need [Git LFS](https://git-lfs.github.com/) to track various datasets - To then launch the jupyter instance use ```bash jupyter notebook ``` -And you should be directed to `localhost:8888`. In the future we should set up the gcloud instance with the same thing. But it seems some bureaucracy got in the way for now. +And you should be directed to `localhost:8888`. ### Project structure * [Models](models) - The resulting generated models @@ -24,7 +21,6 @@ And you should be directed to `localhost:8888`. In the future we should set up t * [Scripts](scripts) - The scripts used for the project * [Datasets](datasets) - The datasets used for the project - ## Resources ### Voice to text @@ -46,23 +42,9 @@ And you should be directed to `localhost:8888`. In the future we should set up t * Adversarial loss [short](https://www.quora.com/What-is-adversarial-loss-in-machine-learning) [paper](https://arxiv.org/pdf/1901.08753.pdf) -### Datasets - -* Conceptual Captions [link](https://ai.google.com/research/ConceptualCaptions/download) - -* Flickr 30k [link](https://www.kaggle.com/hsankesara/flickr-image-dataset/version/1) - -* TIMIT Speech corpus [link](https://catalog.ldc.upenn.edu/LDC93S1) - -I think that we should use the flickr dataset as the 30k images should really be enough in the limited time we have. +* Training GANs, [Tips and Tricks](https://github.com/soumith/ganhacks) -Once you have downloaded the flickr dataset extract it and run the resize script -that's located in the flickr30k_images folder, from the folder in question -```bash -bash resize_images.sh -``` - -## Further notes +## Further information ### Report The report Overleaf is available [here](https://www.overleaf.com/4488118745cjmprgwyfxcw) @@ -72,17 +54,8 @@ The report Overleaf is available [here](https://www.overleaf.com/4488118745cjmpr Might have to use a bag of words model or some other form of context presentation to simplify what the sentence says, look into this further. ### Training GANS -* [Tips and Tricks](https://github.com/soumith/ganhacks) -## GCP -### Running StackGAN -Run StackGAN on GCP from the code folder with -```bash -python2 main.py --cfg cfg/coco_eval.yml --gpu 0 -``` -Contrary to popular belief setting `--gpu 0` here actually refers to the id of the gpu. In most other cases `gpu 0` refers to cpu mode. Weird. - -The generated images will be stored in the `models/coco/netG_epoch_90` directory. +## Google Cloud Platform ### Jupyter notebooks To use jupyter notebooks, run this on the remote @@ -95,6 +68,7 @@ Then tunnel your connection through david@fridge:~$ ssh -N -L localhost:8888:localhost:8888 david@ ``` Then simply open a browser on `localhost:8888` and provide it with the token that should be visible in the commandline window on the vm to connect. + ### Show results The images are viewable in python notebooks and can also, be downloaded from there. diff --git a/datasets/README.md b/datasets/README.md new file mode 100644 index 0000000..1c078c0 --- /dev/null +++ b/datasets/README.md @@ -0,0 +1,15 @@ +# Datasets + +* Conceptual Captions [link](https://ai.google.com/research/ConceptualCaptions/download) + +* Flickr 30k [link](https://www.kaggle.com/hsankesara/flickr-image-dataset/version/1) + +* TIMIT Speech corpus [link](https://catalog.ldc.upenn.edu/LDC93S1) + +I think that we should use the flickr dataset as the 30k images should really be enough in the limited time we have. + +Once you have downloaded the flickr dataset extract it and run the resize script +that's located in the flickr30k_images folder, from the folder in question +```bash +bash resize_images.sh +``` diff --git a/inspiration/README.md b/inspiration/README.md new file mode 100644 index 0000000..9fb732e --- /dev/null +++ b/inspiration/README.md @@ -0,0 +1,8 @@ +## Running StackGAN +Run StackGAN on GCP from the code folder with +```bash +python2 main.py --cfg cfg/coco_eval.yml --gpu 0 +``` +Contrary to popular belief setting `--gpu 0` here actually refers to the id of the gpu. In most other cases `gpu 0` refers to cpu mode. Weird. + +The generated images will be stored in the `models/coco/netG_epoch_90` directory.