Spaces:
Runtime error
Runtime error
| from io import BytesIO | |
| import streamlit as st | |
| import pandas as pd | |
| import os | |
| import numpy as np | |
| from streamlit import caching | |
| from PIL import Image | |
| from model.flax_clip_vision_marian.modeling_clip_vision_marian import ( | |
| FlaxCLIPVisionMarianMT, | |
| ) | |
| from transformers import MarianTokenizer | |
| from utils import ( | |
| get_transformed_image, | |
| ) | |
| import matplotlib.pyplot as plt | |
| from mtranslate import translate | |
| from session import _get_state | |
| state = _get_state() | |
| def load_model(ckpt): | |
| return FlaxCLIPVisionMarianMT.from_pretrained(ckpt) | |
| tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-es") | |
| def generate_sequence(pixel_values, num_beams, temperature, top_p, do_sample, top_k, max_length): | |
| output_ids = state.model.generate(input_ids=pixel_values, max_length=max_length, num_beams=num_beams, temperature=temperature, top_p = top_p, top_k=top_k, do_sample=do_sample) | |
| print(output_ids) | |
| output_sequence = tokenizer.batch_decode(output_ids[0], skip_special_tokens=True, max_length=max_length) | |
| return output_sequence | |
| def read_markdown(path, parent="./sections/"): | |
| with open(os.path.join(parent, path)) as f: | |
| return f.read() | |
| checkpoints = ["./ckpt/ckpt-23999"] # TODO: Maybe add more checkpoints? | |
| dummy_data = pd.read_csv("references.tsv", sep="\t") | |
| st.set_page_config( | |
| page_title="Spanish Image Captioning", | |
| layout="wide", | |
| initial_sidebar_state="collapsed", | |
| page_icon="./misc/csi-logo.png", | |
| ) | |
| st.title("Spanish Image Captioning") | |
| st.write( | |
| "[Bhavitvya Malik](https://huggingface.co/bhavitvyamalik), [Gunjan Chhablani](https://huggingface.co/gchhablani)" | |
| ) | |
| st.sidebar.title("Generation Parameters") | |
| max_length = st.sidebar.number_input("Max Length", min_value=16, max_value=128, value=64, step=1, help="The maximum length of sequence to be generated.") | |
| do_sample = st.sidebar.checkbox("Sample", value=False, help="Sample from the model instead of using beam search.") | |
| top_k = st.sidebar.number_input("Top K", min_value=10, max_value=200, value=50, step=1, help="The number of highest probability vocabulary tokens to keep for top-k-filtering.") | |
| num_beams = st.sidebar.number_input("Number of Beams", min_value=2, max_value=10, value=4, step=1, help="Number of beams to be used in beam search.") | |
| temperature = st.sidebar.select_slider("Temperature", options = list(np.arange(0.0,1.1, step=0.1)), value=1.0, help ="The value used to module the next token probabilities.", format_func=lambda x: f"{x:.2f}") | |
| top_p = st.sidebar.select_slider("Top-P", options = list(np.arange(0.0,1.1, step=0.1)),value=1.0, help="Nucleus Sampling : If set to float < 1, only the most probable tokens with probabilities that add up to :obj:`top_p` or higher are kept for generation.", format_func=lambda x: f"{x:.2f}") | |
| if st.sidebar.button("Clear All Cache"): | |
| caching.clear_cache() | |
| image_col, intro_col = st.beta_columns([3, 8]) | |
| image_col.image("./misc/sic-logo.png", use_column_width="always") | |
| intro_col.write(read_markdown("intro.md")) | |
| with st.beta_expander("Usage"): | |
| st.markdown(read_markdown("usage.md")) | |
| with st.beta_expander("Article"): | |
| st.write(read_markdown("abstract.md")) | |
| st.write(read_markdown("caveats.md")) | |
| st.write("## Methodology") | |
| st.image( | |
| "./misc/Spanish-IC.png" | |
| ) | |
| st.markdown(read_markdown("pretraining.md")) | |
| st.write(read_markdown("challenges.md")) | |
| st.write(read_markdown("social_impact.md")) | |
| st.write(read_markdown("references.md")) | |
| # st.write(read_markdown("checkpoints.md")) | |
| st.write(read_markdown("acknowledgements.md")) | |
| if state.model is None: | |
| with st.spinner("Loading model..."): | |
| state.model = load_model(checkpoints[0]) | |
| first_index = 40 | |
| # Init Session State | |
| if state.image_file is None: | |
| state.image_file = dummy_data.loc[first_index, "image_file"] | |
| state.caption = dummy_data.loc[first_index, "caption"].strip("- ") | |
| image_path = os.path.join("images", state.image_file) | |
| image = plt.imread(image_path) | |
| state.image = image | |
| new_col1, new_col2 = st.beta_columns([5,5]) | |
| if new_col1.button("Get a random example", help="Get a random example from one of the seeded examples."): | |
| sample = dummy_data.sample(1).reset_index() | |
| state.image_file = sample.loc[0, "image_file"] | |
| state.caption = sample.loc[0, "caption"].strip("- ") | |
| image_path = os.path.join("images", state.image_file) | |
| image = plt.imread(image_path) | |
| state.image = image | |
| transformed_image = get_transformed_image(state.image) | |
| # Display Image | |
| new_col1.image(state.image, use_column_width="always") | |
| # Display Reference Caption | |
| with new_col1.beta_expander("Reference Caption"): | |
| st.write("**Reference Caption**: " + state.caption) | |
| st.markdown( | |
| f"""**English Translation**: {translate(state.caption, 'en')}""" | |
| ) | |
| sequence = [''] | |
| if new_col2.button("Generate Caption", help="Generate a caption in the Spanish."): | |
| with st.spinner("Generating Sequence..."): | |
| sequence = generate_sequence(transformed_image, num_beams, temperature, top_p, do_sample, top_k, max_length) | |
| # print(sequence) | |
| if sequence!=['']: | |
| new_col2.write( | |
| "**Generated Caption**: "+sequence[0] | |
| ) | |
| new_col2.write( | |
| "**English Translation**: "+ translate(sequence[0]) | |
| ) | |