Feature list, Standard and Test plan for BETA Release 12/22/2015

===================BETA RELEASE FEATRURE LIST====================

1. Log in and account manager for every user: private for every user.

2. Good UI design and comfortable users' experience: running smoothly and apply for the latest IOS9.

3. Personal photo search: give a txt query (words/sentences) and return the related photos.

4. Personal voice photo search: speech a word or a sentence and return the related photos.

5. Personal photo event segmantation: once you upload your photos, they will be classified according to the event automatically.

6. Personal photo qulity fiter: when you have some photos which is very similar and they contain the same informantion, they will be de-dulicated. If the photos have low quality, they will be removed.

7. Personal photo time and location filter: you can filter your photos according to the time or the GPS information.

8. Process remainder: The process will be displayed and you can check it anytime.

9. Personal photo tagging: the photos will be tagged according to their content automatically.

============================================================

================BEAT RELEASE PERFORMANCE STANDARD================

1. Parallel performance test: The Number of the simultaneous users should be more than 100, and the search result should be return in 3 second.

2. Search performance test: The relevance between the query and the return results accuracy should be more than 60%. Because our CNN model is the AlexNet which the performance upbound is 57.41%.

3. photo quality satisfication:  the score provided by the users according to the How they are satisified with the photo quality. It is divided into 5 ranks. And the user will give the socre of our ALPHA release about the photo quality and de-duplicate feature performance. The final average score result should be more than 4.

4. User experience satisfication: the score provided by the users according to the How they are satisified with the UI design. It is divided into 5 ranks, And the user will give the score of our product about the UI experience. The final average score results should be more than 4.

5. Voice Search test:

1). The voice return words test: for 50 users, let they read some sentence and return words should be hited at least 80%.

2). The NLP extract key words test: the NLP model should extract the key words as the query at leaset 80% when we give the groundtruth.

3). User satisfication test: the score provied by the users according to the degree they feel comfortable when they use the voice search. It is divided into 5 ranks, and the user will give the score. The final average score should more than 4.

============================================================

===================BEAT RELEASE TEST PALN========================

The unit tests will be devided into 4 parts with some test scripts :

1. Search framework test: our search framework is based on the ConSE [1].

we will test the following 3 things:

1). Words coverage rates: give a wordlist and test the hit rate.

2). Stability: whether give some words it will crash or not.

3). Speed: for each query, we will test the return time.

2. NLP mode test: our NLP is based on the stanford API.

we will test the following 2 things:

1). Extract key words accuracy: give a groundtruth and test the hit accuracy.

2). Stability:whether give some words it will crash or not.

3. Voice mode test: our Voice is based on the Oxford API:

we will test the following 2 things:

1). Translation accuracy: users read the sentence and we check the translation from voice sigal to txt accuracy.

2). Stability:whether read some words it will crash or not.

4. Azure server test:

we will deploy our project to the Azure server. The test process will be devided into 3 parts:

1).  Parallel performance test.

2).  loading ability test.

3).  Stability: long time running and no serious bug.

============================================================

Reference:

[1]. M. Norouzi and T. Mikolov. Zero-Shot Learning by Convex Combination of Semantic Embeddings