TypeError: string indices must be integers with model.predict

[carmocca]

🐛 Bug

Traceback (most recent call last):
  File "/Users/carlosmocholi/git/project-flusso/script/kk2.py", line 37, in <module>
    predictions = model.predict(
  File "/Users/carlosmocholi/git/lightning-flash/flash/core/model.py", line 257, in wrapper
    result = func(self, *args, **kwargs)
  File "/Users/carlosmocholi/git/lightning-flash/flash/core/model.py", line 464, in predict
    predictions = self.predict_step(x, 0)  # batch_idx is always 0 when running with `model.predict`
  File "/Users/carlosmocholi/git/lightning-flash/flash/core/adapter.py", line 99, in predict_step
    return self.adapter.predict_step(batch, batch_idx, dataloader_idx=dataloader_idx)
  File "/Users/carlosmocholi/git/lightning-flash/flash/image/classification/adapters.py", line 512, in predict_step
    self._task.task, (batch[DefaultDataKeys.INPUT]), batch_idx, dataloader_idx=dataloader_idx
TypeError: string indices must be integers

Code sample

# Copyright The PyTorch Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch

import flash
from flash.core.data.utils import download_data
from flash.image import ImageClassificationData, ImageClassifier

# 1. Create the DataModule
# Data set from the paper “Movie Genre Classification based on Poster Images with Deep Neural Networks”.
# More info here: https://www.cs.ccu.edu.tw/~wtchu/projects/MoviePoster/
download_data("https://pl-flash-data.s3.amazonaws.com/movie_posters.zip")

datamodule = ImageClassificationData.from_csv(
    "Id",
    ["Action", "Romance", "Crime", "Thriller", "Adventure"],
    train_file="data/movie_posters/train/metadata.csv",
    val_file="data/movie_posters/val/metadata.csv",
    image_size=(128, 128),
)

# 2. Build the task
model = ImageClassifier(backbone="resnet18", num_classes=datamodule.num_classes, multi_label=True)

# 4. Predict the genre of a few movies!
predictions = model.predict(
    [
        "data/movie_posters/predict/tt0085318.jpg",
        "data/movie_posters/predict/tt0089461.jpg",
        "data/movie_posters/predict/tt0097179.jpg",
    ]
)
print(predictions)

Adapted from https://github.com/PyTorchLightning/lightning-flash/blob/master/flash_examples/image_classification_multi_label.py

Expected behavior

Runs

Environment

master

[ethanwharris]

This is a quirk of how the data pipeline is set-up. Until a task is trained / finetuned it doesn't have an attached preprocess (which is then responsible for loading images from files in this case). I think we can set a default preprocess for the task, but ideally we would refactor this in a way that avoids this strange dependency or at least give a warning saying you need to fit or finetune it first.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[ethanwharris]

This is now fixed following the Data Pipeline V2 refactor 😃 All prediction data should be loaded via a datamodule and prediction performed via a Trainer.