fastai--fastai
wehub-resource-sync
561 行
15 KiB
Plaintext
561 行
15 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "raw",
|
|
"id": "f795fba9",
|
|
"metadata": {},
|
|
"source": [
|
|
"---\n",
|
|
"skip_exec: true\n",
|
|
"---"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "8be48eaf",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| default_exp vision.widgets"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "1af69417",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| hide\n",
|
|
"#| eval: false\n",
|
|
"! [ -e /content ] && pip install -Uqq fastai # upgrade fastai on colab"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "25ddc511",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"from fastai.torch_basics import *\n",
|
|
"from fastai.data.all import *\n",
|
|
"from fastai.vision.core import *\n",
|
|
"from fastcore.parallel import *\n",
|
|
"from ipywidgets import HBox,VBox,widgets,Button,Checkbox,Dropdown,Layout,Box,Output,Label,FileUpload"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "32033ab2",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| hide\n",
|
|
"from nbdev.showdoc import *"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "1ed85664",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"_all_ = ['HBox','VBox','widgets','Button','Checkbox','Dropdown','Layout','Box','Output','Label','FileUpload']"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "54a94de7",
|
|
"metadata": {},
|
|
"source": [
|
|
"# Vision widgets\n",
|
|
"\n",
|
|
"> ipywidgets for images"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "f7591688",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"@patch\n",
|
|
"def __getitem__(self:Box, i): return self.children[i]"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "fdc604c5",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"def widget(im, *args, **layout) -> Output:\n",
|
|
" \"Convert anything that can be `display`ed by `IPython` into a widget\"\n",
|
|
" o = Output(layout=merge(*args, layout))\n",
|
|
" with o: display(im)\n",
|
|
" return o"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "fe63a026",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/markdown": [
|
|
"---\n",
|
|
"\n",
|
|
"#### widget\n",
|
|
"\n",
|
|
"> widget (im, *args, **layout)\n",
|
|
"\n",
|
|
"Convert anything that can be `display`ed by `IPython` into a widget"
|
|
],
|
|
"text/plain": [
|
|
"<nbdev.showdoc.BasicMarkdownRenderer>"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"show_doc(widget)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "78ba70e8",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"application/vnd.jupyter.widget-view+json": {
|
|
"model_id": "8c8de2bd3f2d43ff8a00713b47d13782",
|
|
"version_major": 2,
|
|
"version_minor": 0
|
|
},
|
|
"text/plain": [
|
|
"VBox(children=(HTML(value='Puppy'), Output(layout=Layout(max_width='192px'))))"
|
|
]
|
|
},
|
|
"metadata": {},
|
|
"output_type": "display_data"
|
|
}
|
|
],
|
|
"source": [
|
|
"im = Image.open('images/puppy.jpg').to_thumb(256,512)\n",
|
|
"VBox([widgets.HTML('Puppy'),\n",
|
|
" widget(im, max_width=\"192px\")])"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "447f5fde",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"def _update_children(\n",
|
|
" change:dict # A dictionary holding the information about the changed widget\n",
|
|
"):\n",
|
|
" \"Sets a value to the `layout` attribute on widget initialization and change\"\n",
|
|
" for o in change['owner'].children:\n",
|
|
" if not o.layout.flex: o.layout.flex = '0 0 auto'"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "372c2414",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"def carousel(\n",
|
|
" children:tuple|MutableSequence=(), # `Box` objects to display in a carousel\n",
|
|
" **layout\n",
|
|
") -> Box: # An `ipywidget`'s carousel\n",
|
|
" \"A horizontally scrolling carousel\"\n",
|
|
" def_layout = dict(overflow='scroll hidden', flex_flow='row', display='flex')\n",
|
|
" res = Box([], layout=merge(def_layout, layout))\n",
|
|
" res.observe(_update_children, names='children')\n",
|
|
" res.children = children\n",
|
|
" return res"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "ebf07d94",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/markdown": [
|
|
"---\n",
|
|
"\n",
|
|
"#### carousel\n",
|
|
"\n",
|
|
"> carousel (children:Union[tuple,list]=(), **layout)\n",
|
|
"\n",
|
|
"A horizontally scrolling carousel\n",
|
|
"\n",
|
|
"| | **Type** | **Default** | **Details** |\n",
|
|
"| -- | -------- | ----------- | ----------- |\n",
|
|
"| children | tuple \\| list | () | `Box` objects to display in a carousel |\n",
|
|
"| layout | | | |\n",
|
|
"| **Returns** | **Box** | | **An `ipywidget`'s carousel** |"
|
|
],
|
|
"text/plain": [
|
|
"<nbdev.showdoc.BasicMarkdownRenderer>"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"show_doc(carousel)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "ad116929",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"application/vnd.jupyter.widget-view+json": {
|
|
"model_id": "dbaf8c50de1d436ebb7178a5a42673cb",
|
|
"version_major": 2,
|
|
"version_minor": 0
|
|
},
|
|
"text/plain": [
|
|
"Box(children=(VBox(children=(Output(layout=Layout(max_width='192px')), Button(description='click', style=Butto…"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"ts = [VBox([widget(im, max_width='192px'), Button(description='click')])\n",
|
|
" for o in range(3)]\n",
|
|
"\n",
|
|
"carousel(ts, width='450px')"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "2160e34e",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"def _open_thumb(\n",
|
|
" fn:Path|str, # A path of an image\n",
|
|
" h:int, # Thumbnail Height\n",
|
|
" w:int # Thumbnail Width\n",
|
|
") -> Image: # `PIL` image to display\n",
|
|
" \"Opens an image path and returns the thumbnail of the image\"\n",
|
|
" return Image.open(fn).to_thumb(h, w).convert('RGBA')"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "6267f9ee",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"class ImagesCleaner:\n",
|
|
" \"A widget that displays all images in `fns` along with a `Dropdown`\"\n",
|
|
" def __init__(self,\n",
|
|
" opts:tuple=(), # Options for the `Dropdown` menu\n",
|
|
" height:int=128, # Thumbnail Height\n",
|
|
" width:int=256, # Thumbnail Width\n",
|
|
" max_n:int=30 # Max number of images to display\n",
|
|
" ):\n",
|
|
" opts = ('<Keep>', '<Delete>')+tuple(opts)\n",
|
|
" store_attr('opts,height,width,max_n')\n",
|
|
" self.widget = carousel(width='100%')\n",
|
|
"\n",
|
|
" def set_fns(self,\n",
|
|
" fns:list # Contains a path to each image \n",
|
|
" ):\n",
|
|
" \"Sets a `thumbnail` and a `Dropdown` menu for each `VBox`\"\n",
|
|
" self.fns = L(fns)[:self.max_n]\n",
|
|
" ims = parallel(_open_thumb, self.fns, h=self.height, w=self.width, progress=False,\n",
|
|
" n_workers=min(len(self.fns)//10,defaults.cpus))\n",
|
|
" self.widget.children = [VBox([widget(im, height=f'{self.height}px'), Dropdown(\n",
|
|
" options=self.opts, layout={'width': 'max-content'})]) for im in ims]\n",
|
|
"\n",
|
|
" def _ipython_display_(self): display(self.widget)\n",
|
|
" def values(self) -> list:\n",
|
|
" \"Current values of `Dropdown` for each `VBox`\"\n",
|
|
" return L(self.widget.children).itemgot(1).attrgot('value')\n",
|
|
" def delete(self) -> list:\n",
|
|
" \"Indices of items to delete\"\n",
|
|
" return self.values().argwhere(eq('<Delete>'))\n",
|
|
" def change(self) -> list:\n",
|
|
" \"Tuples of the form (index of item to change, new class)\"\n",
|
|
" idxs = self.values().argwhere(not_(in_(['<Delete>','<Keep>'])))\n",
|
|
" return idxs.zipwith(self.values()[idxs])"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "f652c8b5",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/markdown": [
|
|
"---\n",
|
|
"\n",
|
|
"### ImagesCleaner\n",
|
|
"\n",
|
|
"> ImagesCleaner (opts:'tuple'=(), height:'int'=128, width:'int'=256,\n",
|
|
"> max_n:'int'=30)\n",
|
|
"\n",
|
|
"A widget that displays all images in `fns` along with a `Dropdown`\n",
|
|
"\n",
|
|
"| | **Type** | **Default** | **Details** |\n",
|
|
"| -- | -------- | ----------- | ----------- |\n",
|
|
"| opts | tuple | () | Options for the `Dropdown` menu |\n",
|
|
"| height | int | 128 | Thumbnail Height |\n",
|
|
"| width | int | 256 | Thumbnail Width |\n",
|
|
"| max_n | int | 30 | Max number of images to display |"
|
|
],
|
|
"text/plain": [
|
|
"<nbdev.showdoc.BasicMarkdownRenderer>"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"show_doc(ImagesCleaner)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "c15b0eb4",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"application/vnd.jupyter.widget-view+json": {
|
|
"model_id": "deb65e2b8a9342c78df55a8a31186426",
|
|
"version_major": 2,
|
|
"version_minor": 0
|
|
},
|
|
"text/plain": [
|
|
"Box(children=(VBox(children=(Output(layout=Layout(height='128px')), Dropdown(layout=Layout(width='max-content'…"
|
|
]
|
|
},
|
|
"metadata": {},
|
|
"output_type": "display_data"
|
|
}
|
|
],
|
|
"source": [
|
|
"fns = get_image_files('images')\n",
|
|
"w = ImagesCleaner(('A','B'))\n",
|
|
"w.set_fns(fns)\n",
|
|
"w"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "77a69a2a",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/plain": [
|
|
"((#0) [], (#0) [])"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"w.delete(),w.change()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "74875187",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"def _get_iw_info(\n",
|
|
" learn,\n",
|
|
" ds_idx:int=0 # Index in `learn.dls`\n",
|
|
") -> list:\n",
|
|
" \"For every image in `dls` `zip` it's `Path`, target and loss\"\n",
|
|
" dl = learn.dls[ds_idx].new(shuffle=False, drop_last=False)\n",
|
|
" probs,targs,preds,losses = learn.get_preds(dl=dl, with_input=False, with_loss=True, with_decoded=True)\n",
|
|
" targs = [dl.vocab[t] for t in targs]\n",
|
|
" return L([dl.dataset.items,targs,losses]).zip()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "6b5bd73a",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| export\n",
|
|
"@delegates(ImagesCleaner)\n",
|
|
"class ImageClassifierCleaner(GetAttr):\n",
|
|
" \"A widget that provides an `ImagesCleaner` for a CNN `Learner`\"\n",
|
|
" def __init__(self, learn, **kwargs):\n",
|
|
" vocab = learn.dls.vocab\n",
|
|
" self.default = self.iw = ImagesCleaner(vocab, **kwargs)\n",
|
|
" self.dd_cats = Dropdown(options=vocab)\n",
|
|
" self.dd_ds = Dropdown(options=('Train','Valid'))\n",
|
|
" self.iwis = _get_iw_info(learn,0),_get_iw_info(learn,1)\n",
|
|
" self.dd_ds.observe(self.on_change_ds, 'value')\n",
|
|
" self.dd_cats.observe(self.on_change_ds, 'value')\n",
|
|
" self.on_change_ds()\n",
|
|
" self.widget = VBox([self.dd_cats, self.dd_ds, self.iw.widget])\n",
|
|
"\n",
|
|
" def _ipython_display_(self): display(self.widget)\n",
|
|
" def on_change_ds(self,change=None):\n",
|
|
" \"Toggle between training validation set view\"\n",
|
|
" info = L(o for o in self.iwis[self.dd_ds.index] if o[1]==self.dd_cats.value)\n",
|
|
" self.iw.set_fns(info.sorted(2, reverse=True).itemgot(0))"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "808b4e9a",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/markdown": [
|
|
"---\n",
|
|
"\n",
|
|
"[source](https://github.com/fastai/fastai/blob/main/fastai/vision/widgets.py#L108){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
|
|
"\n",
|
|
"### ImageClassifierCleaner\n",
|
|
"\n",
|
|
"> ImageClassifierCleaner (learn, opts:tuple=(), height:int=128,\n",
|
|
"> width:int=256, max_n:int=30)\n",
|
|
"\n",
|
|
"*A widget that provides an `ImagesCleaner` for a CNN `Learner`*\n",
|
|
"\n",
|
|
"| | **Type** | **Default** | **Details** |\n",
|
|
"| -- | -------- | ----------- | ----------- |\n",
|
|
"| learn | | | |\n",
|
|
"| opts | tuple | () | Options for the `Dropdown` menu |\n",
|
|
"| height | int | 128 | Thumbnail Height |\n",
|
|
"| width | int | 256 | Thumbnail Width |\n",
|
|
"| max_n | int | 30 | Max number of images to display |"
|
|
],
|
|
"text/plain": [
|
|
"---\n",
|
|
"\n",
|
|
"[source](https://github.com/fastai/fastai/blob/main/fastai/vision/widgets.py#L108){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
|
|
"\n",
|
|
"### ImageClassifierCleaner\n",
|
|
"\n",
|
|
"> ImageClassifierCleaner (learn, opts:tuple=(), height:int=128,\n",
|
|
"> width:int=256, max_n:int=30)\n",
|
|
"\n",
|
|
"*A widget that provides an `ImagesCleaner` for a CNN `Learner`*\n",
|
|
"\n",
|
|
"| | **Type** | **Default** | **Details** |\n",
|
|
"| -- | -------- | ----------- | ----------- |\n",
|
|
"| learn | | | |\n",
|
|
"| opts | tuple | () | Options for the `Dropdown` menu |\n",
|
|
"| height | int | 128 | Thumbnail Height |\n",
|
|
"| width | int | 256 | Thumbnail Width |\n",
|
|
"| max_n | int | 30 | Max number of images to display |"
|
|
]
|
|
},
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"show_doc(ImageClassifierCleaner)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "72fe3775",
|
|
"metadata": {},
|
|
"source": [
|
|
"# Export -"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "9f6cbc7d",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"#| hide\n",
|
|
"import nbdev; nbdev.nbdev_export()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "2f3fda3d",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": []
|
|
}
|
|
],
|
|
"metadata": {
|
|
"jupytext": {
|
|
"split_at_heading": true
|
|
},
|
|
"kernelspec": {
|
|
"display_name": "python3",
|
|
"language": "python",
|
|
"name": "python3"
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 5
|
|
}
|