Spaces:
Running
Running
File size: 5,478 Bytes
6d787c4 c670c44 6d787c4 c670c44 6d787c4 c670c44 6d787c4 11296d8 |
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 |
from datasets import load_dataset, IterableDataset
from functools import partial
from pandas import DataFrame
import gradio as gr
import numpy as np
import tqdm
import json
import os
DEBUG = False
sets = {
"satellogic": {
"shards" : 3676,
},
"sentinel_1": {
"shards" : 1763,
},
"neon": {
"config" : "default",
"shards" : 607,
"path" : "data",
}
}
def open_dataset(dataset, set_name, split, batch_size, shard = -1):
global dsi, ds
if shard == -1:
data_files = None
shards = 100
else:
config = sets[set_name].get("config", set_name)
shards = sets[set_name]["shards"]
path = sets[set_name].get("path", set_name)
data_files = {"train":[f"{path}/{split}-{shard:05d}-of-{shards:05d}.parquet"]}
if DEBUG:
ds = lambda:None
ds.n_shards = 1234
dsi = range(100)
else:
ds = load_dataset(
dataset,
config,
split=split,
cache_dir="dataset",
data_files=data_files,
streaming=True,
use_auth_token=os.environ.get("HF_TOKEN"))
dsi = iter(ds)
return (
gr.update(label=f"Shards (max {shards})", value=shard, maximum=shards),
*get_images(batch_size)
)
def get_images(batch_size):
global dsi
items = []
metadatas = []
for i in tqdm.trange(batch_size, desc=f"Getting images"):
if DEBUG:
image = np.random.randint(0,255,(384,384,3))
metadata = {"bounds":[[1,1,4,4]], }
else:
try:
item = next(dsi)
except StopIteration:
break
metadata = item["metadata"]
if ds.config_name == "satellogic":
image = np.asarray(item["rgb"][0]).astype(np.uint8)
items.append(image.transpose(1,2,0))
if ds.config_name == "sentinel_1":
metadata = json.loads(metadata)
data = np.asarray(item["10m"])
for i in range(data.shape[0]):
# Mapping of V and H to RGB. May not be correct
# https://gis.stackexchange.com/questions/400726/creating-composite-rgb-images-from-sentinel-1-channels
image = np.zeros((3,384,384), "uint8")
image[0] = data[i][0]
image[1] = data[i][1]
image[2] = (image[0]/(image[1]+0.1))*256
items.append(image.transpose(1,2,0))
if ds.config_name == "default":
dataRGB = np.asarray(item["rgb"]).astype("uint8")
dataCHM = np.asarray(item["chm"]).astype("uint8")
data1m = np.asarray(item["1m"]).astype("uint8")
for i in range(dataRGB.shape[0]):
image = dataRGB[i,:,:,:]
items.append(image.transpose(1,2,0))
image = dataCHM[i,0,:,:]
items.append(image)
image = data1m[i,0,:,:]
items.append(image)
metadatas.append(metadata)
return items, DataFrame(metadatas)
def skip(count, batch_size):
global dsi
skip = count*batch_size
gr.Info(f"Skipping {skip} images (it's slow)")
for i in tqdm.trange(skip, desc=f"Skipping {skip} images"):
if DEBUG:
pass
else:
next(dsi)
return get_images(batch_size)
def update_shape(rows, columns):
return gr.update(rows=rows, columns=columns)
with gr.Blocks(title="Dataset Explorer", fill_height = True) as demo:
batch_size = gr.Number(10, label = "Batch Size", render=False)
shard = gr.Slider(label="Shard", minimum=0, maximum=10000, step=1, render=False)
table = gr.DataFrame(render = False)
# headers=["Index","TimeStamp","Bounds","CRS"],
gallery = gr.Gallery(
label="satellogic/EarthView",
interactive=False,
columns=5, rows=2, render=False)
with gr.Row():
dataset = gr.Textbox(label="Dataset", value="satellogic/EarthView")
config = gr.Dropdown(choices=["satellogic", "sentinel_1", "neon"], label="Subset", value="satellogic", )
split = gr.Textbox(label="Split", value="train")
initial_shard = gr.Number(label = "Initial shard", value=0)
gr.Button("Load (minutes)").click(
open_dataset,
inputs=[dataset, config, split, batch_size, initial_shard],
outputs=[shard, gallery, table])
gallery.render()
with gr.Row():
batch_size.render()
rows = gr.Number(2, label="Rows")
columns = gr.Number(5, label="Coluns")
rows.change(update_shape, [rows, columns], [gallery])
columns.change(update_shape, [rows, columns], [gallery])
with gr.Row():
shard.render()
shard.release(
open_dataset,
inputs=[dataset, config, split, batch_size, shard],
outputs=[shard, gallery, table])
btn = gr.Button("Get More Images", scale=0)
btn.click(get_images, [batch_size], [gallery, table])
btn.click()
# btn = gr.Button("Skip 10 Batches", scale=0)
# btn.click(partial(skip, 10), [batch], gallery)
# btn = gr.Button("Skip 25 Batches", scale=0)
# btn.click(partial(skip, 25), [batch], gallery)
table.render()
demo.launch(show_api=False)
|