Spaces:
Runtime error
Runtime error
File size: 6,153 Bytes
eea614c 268ddb9 eea614c 268ddb9 eea614c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 |
import gradio as gr
import numpy as np
import torch
import jax
import jax.numpy as jnp
from flax.jax_utils import replicate
from flax.training.common_utils import shard
from PIL import Image
from segment_anything import SamPredictor, sam_model_registry, SamAutomaticMaskGenerator
from diffusers import (
UniPCMultistepScheduler,
FlaxStableDiffusionControlNetPipeline,
FlaxControlNetModel,
)
import colorsys
sam_checkpoint = "sam_vit_h_4b8939.pth"
model_type = "vit_h"
device = "cpu"
sam = sam_model_registry[model_type](checkpoint=sam_checkpoint)
sam.to(device=device)
predictor = SamPredictor(sam)
mask_generator = SamAutomaticMaskGenerator(sam)
controlnet, controlnet_params = FlaxControlNetModel.from_pretrained(
"SAMControlNet/sd-controlnet-sam-seg", dtype=jnp.float32
)
pipe, params = FlaxStableDiffusionControlNetPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
controlnet=controlnet,
revision="flax",
dtype=jnp.bfloat16,
)
params["controlnet"] = controlnet_params
p_params = replicate(params)
with gr.Blocks() as demo:
gr.Markdown("# WildSynth: Synthetic Wildlife Data Generation")
gr.Markdown(
"""
### About
We have trained a JAX ControlNet model for semantic segmentation on Wildlife Animal Images.
For the training data creation we used the [Wildlife Animals Images](https://www.kaggle.com/datasets/anshulmehtakaggl/wildlife-animals-images) dataset.
We created segmentation masks with the help of [Grounded SAM](https://github.com/IDEA-Research/Grounded-Segment-Anything) where we used the animals names
as input prompts for detection and more accurate segmentation.
### How To Use
"""
)
with gr.Row():
input_img = gr.Image(label="Input")
mask_img = gr.Image(label="Mask", interactive=False)
output_img = gr.Image(label="Output", interactive=False)
with gr.Row():
prompt_text = gr.Textbox(lines=1, label="Prompt")
negative_prompt_text = gr.Textbox(lines=1, label="Negative Prompt")
with gr.Row():
submit = gr.Button("Submit")
clear = gr.Button("Clear")
def generate_mask(image, evt: gr.SelectData):
predictor.set_image(image)
input_point = np.array([120, 21])
input_label = np.ones(input_point.shape[0])
mask, _, _ = predictor.predict(
point_coords=input_point,
point_labels=input_label,
multimask_output=False,
)
# clear torch cache
torch.cuda.empty_cache()
mask = Image.fromarray(mask[0, :, :])
segs = mask_generator.generate(image)
boolean_masks = [s["segmentation"] for s in segs]
finseg = np.zeros(
(boolean_masks[0].shape[0], boolean_masks[0].shape[1], 3), dtype=np.uint8
)
# Loop over the boolean masks and assign a unique color to each class
for class_id, boolean_mask in enumerate(boolean_masks):
hue = class_id * 1.0 / len(boolean_masks)
rgb = tuple(int(i * 255) for i in colorsys.hsv_to_rgb(hue, 1, 1))
rgb_mask = np.zeros(
(boolean_mask.shape[0], boolean_mask.shape[1], 3), dtype=np.uint8
)
rgb_mask[:, :, 0] = boolean_mask * rgb[0]
rgb_mask[:, :, 1] = boolean_mask * rgb[1]
rgb_mask[:, :, 2] = boolean_mask * rgb[2]
finseg += rgb_mask
torch.cuda.empty_cache()
return mask, finseg
def infer(
image, prompts, negative_prompts, num_inference_steps=50, seed=4, num_samples=4
):
try:
rng = jax.random.PRNGKey(int(seed))
num_inference_steps = int(num_inference_steps)
image = Image.fromarray(image, mode="RGB")
num_samples = max(jax.device_count(), int(num_samples))
p_rng = jax.random.split(rng, jax.device_count())
prompt_ids = pipe.prepare_text_inputs([prompts] * num_samples)
negative_prompt_ids = pipe.prepare_text_inputs(
[negative_prompts] * num_samples
)
processed_image = pipe.prepare_image_inputs([image] * num_samples)
prompt_ids = shard(prompt_ids)
negative_prompt_ids = shard(negative_prompt_ids)
processed_image = shard(processed_image)
output = pipe(
prompt_ids=prompt_ids,
image=processed_image,
params=p_params,
prng_seed=p_rng,
num_inference_steps=num_inference_steps,
neg_prompt_ids=negative_prompt_ids,
jit=True,
).images
del negative_prompt_ids
del processed_image
del prompt_ids
output = output.reshape((num_samples,) + output.shape[-3:])
final_image = [np.array(x * 255, dtype=np.uint8) for x in output]
print(output.shape)
del output
except Exception as e:
print("Error: " + str(e))
final_image = [np.zeros((512, 512, 3), dtype=np.uint8)] * num_samples
finally:
gc.collect()
return final_image
def _clear(sel_pix, img, mask, seg, out, prompt, neg_prompt, bg):
img = None
mask = None
seg = None
out = None
prompt = ""
neg_prompt = ""
bg = False
return img, mask, seg, out, prompt, neg_prompt, bg
input_img.change(
generate_mask,
inputs=[input_img],
outputs=[mask_img],
)
submit.click(
infer,
inputs=[mask_img, prompt_text, negative_prompt_text],
outputs=[output_img],
)
clear.click(
_clear,
inputs=[
input_img,
mask_img,
output_img,
prompt_text,
negative_prompt_text,
],
outputs=[
input_img,
mask_img,
output_img,
prompt_text,
negative_prompt_text,
],
)
if __name__ == "__main__":
demo.queue()
demo.launch()
|