madhavan001 / image-captioning-approaches Goto Github PK

1. Take the most frequently used words from the corpus

1. Convert to Lower case
2. Remove punctuations.
3. Remove empty tokens

Evaluate

1. Top k index
2. Sample Images vs mis-labelled captions
3. Blew Graphs

Dataset:

1. Describe the dataset
2. Sample Images vs captions

Use Batch Normalization and Dropout and compare the results

The current implementation of one-time encoding using cnn takes a long time. Improve this by multi-threading the process

Implement transfer learning for the language model.

Currently, we are learning the model from scratch conditioned on the image

Slide Structure:

1. Introduction
2. Literature Review
3. Architecture
4. Dataset + Cleaning Procedure for our dataset
5. Tasks Completed
6. Pending Tasks
7. Preliminary Analysis of Dataset
8. Pending Analysis (Completed + Planning)

Replace this with Google Inception Model Used in the existing job

Considerations:

1. Flask
2. Python Django

Reduce the size of the dataset of varying sizes 10k, 100k and Full Images vs Captions

Steps:

1. Clean the dataset
2. Write the scripts to directly download the dataset to the virtual machine
3. Run the script and populate the dataset in GCP VM

This needs to be added to language/decoder.py