diff --git a/.gitignore b/.gitignore
index f9058f7..9f03192 100644
--- a/.gitignore
+++ b/.gitignore
@@ -89,11 +89,17 @@ ENV/
 .ropeproject
 
 
-# dataset related resources
+# project structure related resources
 data/*
+experiments/*
+pretrained/*
+
 # do not ignore dataset downloader
-!data/download_mscoco.sh
+!data/README.md
+!experiments/README.md
+!pretrained/README.md
 *.h5
 
 # ignore config
 *.json
+
diff --git a/config/README.md b/config/README.md
new file mode 100644
index 0000000..fcfa66e
--- /dev/null
+++ b/config/README.md
@@ -0,0 +1,5 @@
+# Configuration files
+
+This is the directory where the configuration for the experiments will be loaded from.
+
+In order to run a custom experiment, copy the `solver.json.default` file into any other name, e.g. `train_enet.json` and run `python src/run.py ${mode} train_enet.json` from the root directory of the project. `mode` can only take the value of `train` at the moment. `predict` and `test`/`evaluate` functionality is planned, as well as a `python run.py --help` menu.
diff --git a/data/README.md b/data/README.md
new file mode 100644
index 0000000..6aa626a
--- /dev/null
+++ b/data/README.md
@@ -0,0 +1,5 @@
+# Data directory
+
+Placeholder.
+
+TODO: add contents
diff --git a/data/download_mscoco.sh b/data/download_mscoco.sh
deleted file mode 100755
index 3fd5193..0000000
--- a/data/download_mscoco.sh
+++ /dev/null
@@ -1,8 +0,0 @@
-# train set
-wget -P mscoco http://msvocds.blob.core.windows.net/coco2014/train2014.zip
-
-# validation set
-wget -P mscoco http://msvocds.blob.core.windows.net/coco2014/val2014.zip
-
-# annotations
-wget -P mscoco http://msvocds.blob.core.windows.net/annotations-1-0-3/instances_train-val2014.zip
diff --git a/experiments/README.md b/experiments/README.md
new file mode 100644
index 0000000..0cd6a30
--- /dev/null
+++ b/experiments/README.md
@@ -0,0 +1,4 @@
+# Experiments directory
+
+This is the directory where the tensorboard files (under `logs/`) as well as the model snapshots (under `weights/`) will be stored.
+The directory will be populated automatically. The directory structure is experiments/${dataset_name}/${model_name}/${[logs|weights]}
diff --git a/pretrained/README.md b/pretrained/README.md
new file mode 100644
index 0000000..b138a88
--- /dev/null
+++ b/pretrained/README.md
@@ -0,0 +1,5 @@
+# Pretrained model directory
+
+This is the directory where the pretrained models will be loaded from.
+
+Run `./setup.sh` to download the enet model and convert it to a keras-compatible model before you run the actual training/prediction scripts.