Image2Text Dashboard

Dashboard for Experiments
Created on March 2|Last edited on April 12
Comment
﻿
Training 🛠
Data 💾The data input for training is 
an image whose features are extracted via InceptionV3. 
caption which is converted into a word vector with index = <position in sentence> and value = <id of word as defined by the constructed vocabulary>
We show here the original images alongside the caption and the associated relative file name as defined in our artifact structure
For ✨flair✨ I added an interactive way to view the images based on the word vectors of the caption (may not make the most sense but it's fun!)
﻿
﻿
project("fauxvo-faux", "image_captioning").artifactVersion("split", "cd97f3b6eb6382aef737").file("train_img_cap_table.table.json")
 - 2 of 4800
name.caption
image
1
COCO_train2014_000000321666.jpg
2
COCO_train2014_000000460039.jpg
name.name
﻿
project("fauxvo-faux", "image_captioning").artifactVersion("split", "cd97f3b6eb6382aef737").file("train_img_cap_table.table.json")
Losses 📉﻿
epoch_loss
epoch_loss
2k4k6kStep0.40.60.81
train-coco2014-attention-model
batch_loss
batch_loss
02k4k6kStep20406080100
train-coco2014-attention-model
Run: train-coco2014-attention-model1
﻿
﻿
Gradients 🎱
Evaluation 📸→✍🏽
Predictions 🙌🏽For each prediction, we also include the attention plot to allow us to better interpret why the model may be making the predictions it made. Take a look! 🔍
💡
Logging this as a table makes it very easy for the ✍🏽→🗣 team to pull and run their inference on!
﻿
﻿
project("fauxvo-faux", "image_captioning").artifactVersion("eval_table", "8eb982e404b2805a7bb1").file("eval_table.table.json")
 - 22 of 1200
image
real_caption
predicted_caption
attention_plot
1
COCO_train2014_000000397526.jpg
2
COCO_train2014_000000480861.jpg
name
﻿
Add a comment