How to run run text to image on MS-COCO database

MoslemTCM commented 6 years ago

Description

Hi, i am trying to run the Image2Text problem (image_text_ms_coco) using Tensor2Tensor on MS-COCO database. But I'm not sure what set of parameter (problem, model and hparams) should i use ? Should i change these settings if i want to test the opposite problem text2image (image_text_ms_coco_rev).

Environment information

OS: Ubuntu 16.04

$ pip freeze | grep tensor
tensor2tensor==1.8.0
tensorboard==1.10.0
tensorflow-gpu==1.10.0

$ python -V
Python 3.5.2

For bugs: reproduction and error logs

# Steps to reproduce:
HPARAMS=imagetransformer2d_base
PROBLEM=image_text_ms_coco
MODEL=imagetransformer2d
DATA_DIR=/opt/T2T_Image_Caption/t2t_data/
TRAIN_DIR=/opt/T2T_Image_Caption/t2t_train/image_text_ms_coco/

t2t-trainer \
  --data_dir=$DATA_DIR \
  --output_dir=$TRAIN_DIR \
  --train_steps=4000000 \
  --model=$MODEL \
  --hparams_set=$HPARAMS \
  --problem=$PROBLEM

# Error logs:
ValueError: Dimension size must be evenly divisible by 3 but is 1
    Number of ways to split should evenly divide the split dimension for 'imagetransformer2d/parallel_0_4/imagetransformer2d/imagetransformer2d/body/split' (op: 'Split') with input shapes: [], [?,?,1,1] and with computed input tensors: input[0] = <3>.

ZainySong commented 5 years ago

So sorry to bother you. I met the same problem as you. Have you solved this problem?

MoslemTCM commented 5 years ago

@ZainySong , sorry but i am still looking for a solution.

deeptimhe commented 5 years ago

@MoslemTCM @ZainySong I am working on this too. Have you solved this?

tensorflow / tensor2tensor