from gradio.outputs import Label
from icevision import tfms
from icevision.models.checkpoint import model_from_checkpoint
import PIL
import gradio as gr
import os

detection_threshold = 0.7
checkpoint_path = "2022-10-03_resnet_slates_147.pth"
description = (
    "A faster-rcnn model that detects film slates / clappers. "
    "Upload an image of a slate or click an example below!"
)

# Load model
checkpoint_and_model = model_from_checkpoint(checkpoint_path)
model = checkpoint_and_model["model"]
model_type = checkpoint_and_model["model_type"]
class_map = checkpoint_and_model["class_map"]

# Transforms
img_size = checkpoint_and_model["img_size"]
valid_tfms = tfms.A.Adapter(
    [*tfms.A.resize_and_pad(img_size), tfms.A.Normalize()]
)

# Populate examples in Gradio interface
examples = [
    ["1.jpg"],
    ["2.jpg"],
    ["3.jpg"],
]


def show_preds(input_image):
    img = PIL.Image.fromarray(input_image, "RGB")
    pred_dict = model_type.end2end_detect(
        img,
        valid_tfms,
        model,
        class_map=class_map,
        detection_threshold=detection_threshold,
        display_label=False,
        display_bbox=True,
        return_img=True,
        font_size=16,
        label_color="#02D800",
    )
    return pred_dict["img"]


gr_interface = gr.Interface(
    fn=show_preds,
    inputs=["image"],
    outputs=[gr.outputs.Image(type="pil", label="Inference")],
    title="Film Slate Detector",
    description=description,
    examples=examples,
)
gr_interface.launch(inline=False, share=False, debug=True)