Turn on FAISS by default and use FAISS flat retrievers.

Soon: Export trained approximators. FAISS K-NN graph.

Next:

- Implement pre-condition checks, failed to launch error
messages for processes. Rollback created models on failure to launch.

- Implement clustering.

- Implement location net, fine-tuning / training face recognition models

.travis.yml

@@ -20,7 +20,6 @@ virtualenv:
 env:
   global:
     - AUTH_DISABLED=1
-    - ENABLE_FAISS=1
     - SECRET_KEY=123213123123213213
     - LAUNCH_BY_NAME_indexer_inception=1
     - LAUNCH_BY_NAME_retriever_inception=1

deploy/compose/dev/docker-compose.yml

@@ -41,7 +41,6 @@
        - LAUNCH_BY_NAME_detector_textbox=1
        - LAUNCH_Q_qextract=1
        - LAUNCH_Q_qtrainer=1
-       - ENABLE_FAISS=1
        - LAUNCH_Q_qstreamer=1
        - LAUNCH_SCHEDULER=1
        - LAUNCH_SERVER=1 # This supersedes NGINX and launches django development server with auto-reload

server/dva/settings.py

@@ -295,5 +295,5 @@ def show_toolbar(request):
 # Max task attempts
 MAX_TASK_ATTEMPTS = 5
 # FAISS
-ENABLE_FAISS = 'ENABLE_FAISS' in os.environ
+ENABLE_FAISS = 'DISABLE_FAISS' not in os.environ