Spaces:
Running
Running
File size: 13,827 Bytes
a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 0aab987 a11eff5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 |
{
"cells": [
{
"cell_type": "markdown",
"id": "f6d33374",
"metadata": {},
"source": [
"# Test notebook with CLIP scoring"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6eb74941-bb4d-4d7e-97f1-d5a3a07672bf",
"metadata": {},
"outputs": [],
"source": [
"# !pip install flax transformers\n",
"# !git clone https://github.com/patil-suraj/vqgan-jax.git"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "41db7534-f589-4b63-9165-9c9799e1b06e",
"metadata": {},
"outputs": [],
"source": [
"import random\n",
"\n",
"import jax\n",
"import flax.linen as nn\n",
"from flax.training.common_utils import shard\n",
"from flax.jax_utils import replicate, unreplicate\n",
"\n",
"from transformers.models.bart.modeling_flax_bart import *\n",
"from transformers import BartTokenizer, FlaxBartForConditionalGeneration\n",
"\n",
"import io\n",
"\n",
"import requests\n",
"from PIL import Image\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"\n",
"import torch\n",
"import torchvision.transforms as T\n",
"import torchvision.transforms.functional as TF\n",
"from torchvision.transforms import InterpolationMode\n",
"\n",
"jax.devices()"
]
},
{
"cell_type": "markdown",
"id": "d408065c",
"metadata": {},
"source": [
"`dalle_mini` is a local package that contains the VQGAN-JAX model by Suraj, and other utilities. You can also `cd` to the directory that contains your checkout of [`vqgan-jax`](https://github.com/patil-suraj/vqgan-jax.git)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "09295910",
"metadata": {},
"outputs": [],
"source": [
"from dalle_mini.vqgan_jax.modeling_flax_vqgan import VQModel\n",
"#%cd /content/vqgan-jax"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b6a3462a-9004-4121-b365-3ae3aaf94dd2",
"metadata": {},
"outputs": [],
"source": [
"# TODO: set those args in a config file\n",
"OUTPUT_VOCAB_SIZE = 16384 + 1 # encoded image token space + 1 for bos\n",
"OUTPUT_LENGTH = 256 + 1 # number of encoded tokens + 1 for bos\n",
"BOS_TOKEN_ID = 16384\n",
"BASE_MODEL = 'facebook/bart-large-cnn'"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bbef1afb-0b36-44a5-83f7-643d7e2c0e30",
"metadata": {},
"outputs": [],
"source": [
"class CustomFlaxBartModule(FlaxBartModule):\n",
" def setup(self):\n",
" # we keep shared to easily load pre-trained weights\n",
" self.shared = nn.Embed(\n",
" self.config.vocab_size,\n",
" self.config.d_model,\n",
" embedding_init=jax.nn.initializers.normal(self.config.init_std, self.dtype),\n",
" dtype=self.dtype,\n",
" )\n",
" # a separate embedding is used for the decoder\n",
" self.decoder_embed = nn.Embed(\n",
" OUTPUT_VOCAB_SIZE,\n",
" self.config.d_model,\n",
" embedding_init=jax.nn.initializers.normal(self.config.init_std, self.dtype),\n",
" dtype=self.dtype,\n",
" )\n",
" self.encoder = FlaxBartEncoder(self.config, dtype=self.dtype, embed_tokens=self.shared)\n",
"\n",
" # the decoder has a different config\n",
" decoder_config = BartConfig(self.config.to_dict())\n",
" decoder_config.max_position_embeddings = OUTPUT_LENGTH\n",
" decoder_config.vocab_size = OUTPUT_VOCAB_SIZE\n",
" self.decoder = FlaxBartDecoder(decoder_config, dtype=self.dtype, embed_tokens=self.decoder_embed)\n",
"\n",
"class CustomFlaxBartForConditionalGenerationModule(FlaxBartForConditionalGenerationModule):\n",
" def setup(self):\n",
" self.model = CustomFlaxBartModule(config=self.config, dtype=self.dtype)\n",
" self.lm_head = nn.Dense(\n",
" OUTPUT_VOCAB_SIZE,\n",
" use_bias=False,\n",
" dtype=self.dtype,\n",
" kernel_init=jax.nn.initializers.normal(self.config.init_std, self.dtype),\n",
" )\n",
" self.final_logits_bias = self.param(\"final_logits_bias\", self.bias_init, (1, OUTPUT_VOCAB_SIZE))\n",
"\n",
"class CustomFlaxBartForConditionalGeneration(FlaxBartForConditionalGeneration):\n",
" module_class = CustomFlaxBartForConditionalGenerationModule"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "879320b7-eaa0-4dc9-bbf2-c81efc53301d",
"metadata": {},
"outputs": [],
"source": [
"import wandb\n",
"run = wandb.init()\n",
"artifact = run.use_artifact('wandb/hf-flax-dalle-mini/model-3h3x3565:latest', type='bart_model')\n",
"artifact_dir = artifact.download()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e8bcff33-e95b-4c01-b162-ee857a55c3e6",
"metadata": {},
"outputs": [],
"source": [
"# create our model\n",
"tokenizer = BartTokenizer.from_pretrained(BASE_MODEL)\n",
"model = CustomFlaxBartForConditionalGeneration.from_pretrained(artifact_dir)\n",
"model.config.force_bos_token_to_be_generated = False\n",
"model.config.forced_bos_token_id = None\n",
"model.config.forced_eos_token_id = None\n",
"\n",
"# we verify that the shape has not been modified\n",
"model.params['final_logits_bias'].shape"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8d5e0f14-2502-470e-9553-daee6748601f",
"metadata": {},
"outputs": [],
"source": [
"vqgan = VQModel.from_pretrained(\"flax-community/vqgan_f16_16384\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6cca395a-93c2-49bc-a3be-98287e4403d4",
"metadata": {},
"outputs": [],
"source": [
"def custom_to_pil(x):\n",
" x = np.clip(x, 0., 1.)\n",
" x = (255*x).astype(np.uint8)\n",
" x = Image.fromarray(x)\n",
" if not x.mode == \"RGB\":\n",
" x = x.convert(\"RGB\")\n",
" return x\n",
"\n",
"def generate(input, rng, params):\n",
" return model.generate(\n",
" **input,\n",
" max_length=257,\n",
" num_beams=1,\n",
" do_sample=True,\n",
" prng_key=rng,\n",
" eos_token_id=50000,\n",
" pad_token_id=50000,\n",
" params=params\n",
" )\n",
"\n",
"def get_images(indices, params):\n",
" return vqgan.decode_code(indices, params=params)\n",
"\n",
"\n",
"def plot_images(images):\n",
" fig = plt.figure(figsize=(40, 20))\n",
" columns = 4\n",
" rows = 2\n",
" plt.subplots_adjust(hspace=0, wspace=0)\n",
"\n",
" for i in range(1, columns*rows +1):\n",
" fig.add_subplot(rows, columns, i)\n",
" plt.imshow(images[i-1])\n",
" plt.gca().axes.get_yaxis().set_visible(False)\n",
" plt.show()\n",
" \n",
"def stack_reconstructions(images):\n",
" w, h = images[0].size[0], images[0].size[1]\n",
" img = Image.new(\"RGB\", (len(images)*w, h))\n",
" for i, img_ in enumerate(images):\n",
" img.paste(img_, (i*w,0))\n",
" return img"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b1bec3d2-ef17-4feb-aa0d-b51ed2fdcd3e",
"metadata": {},
"outputs": [],
"source": [
"p_generate = jax.pmap(generate, \"batch\")\n",
"p_get_images = jax.pmap(get_images, \"batch\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a539823a-a775-4d92-96a5-dc8b1eef69c5",
"metadata": {},
"outputs": [],
"source": [
"bart_params = replicate(model.params)\n",
"vqgan_params = replicate(vqgan.params)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e8b268d8-6992-422a-8373-95651474ae70",
"metadata": {},
"outputs": [],
"source": [
"prompts = [\n",
" \"man in blue jacket walking on pathway in between trees during daytime\",\n",
" 'white snow covered mountain under blue sky during daytime',\n",
" 'white snow covered mountain under blue sky during night',\n",
" \"orange tabby cat on persons hand\",\n",
" \"aerial view of beach during daytime\",\n",
" \"chess pieces on chess board\",\n",
" \"laptop on brown wooden table\",\n",
" \"white bus on road near high rise buildings\",\n",
"]\n",
"\n",
"\n",
"prompt = [prompts[1]] * jax.device_count()\n",
"inputs = tokenizer(prompt, return_tensors='jax', padding=\"max_length\", truncation=True, max_length=128).data\n",
"inputs = shard(inputs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "68638cfa-9a4d-4e6a-8630-91aefb627bbd",
"metadata": {},
"outputs": [],
"source": [
"%%time\n",
"for i in range(8):\n",
" key = random.randint(0, 1e7)\n",
" rng = jax.random.PRNGKey(key)\n",
" rngs = jax.random.split(rng, jax.local_device_count())\n",
" indices = p_generate(inputs, rngs, bart_params).sequences\n",
" indices = indices[:, :, 1:]\n",
"\n",
" images = p_get_images(indices, vqgan_params)\n",
" images = np.squeeze(np.asarray(images), 1)\n",
" imges = [custom_to_pil(image) for image in images]\n",
"\n",
" plt.figure(figsize=(40, 20))\n",
" plt.imshow(stack_reconstructions(imges))"
]
},
{
"cell_type": "markdown",
"id": "b6e1060f",
"metadata": {},
"source": [
"## CLIP Scoring"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c68724bc",
"metadata": {},
"outputs": [],
"source": [
"from transformers import CLIPProcessor, FlaxCLIPModel"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "17158e5b",
"metadata": {},
"outputs": [],
"source": [
"clip = FlaxCLIPModel.from_pretrained(\"openai/clip-vit-base-patch32\")\n",
"processor = CLIPProcessor.from_pretrained(\"openai/clip-vit-base-patch32\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1b37b6d",
"metadata": {},
"outputs": [],
"source": [
"def hallucinate(prompt, num_images=64):\n",
" prompt = [prompt] * jax.device_count()\n",
" inputs = tokenizer(prompt, return_tensors='jax', padding=\"max_length\", truncation=True, max_length=128).data\n",
" inputs = shard(inputs)\n",
"\n",
" all_images = []\n",
" for i in range(num_images // jax.device_count()):\n",
" key = random.randint(0, 1e7)\n",
" rng = jax.random.PRNGKey(key)\n",
" rngs = jax.random.split(rng, jax.local_device_count())\n",
" indices = p_generate(inputs, rngs, bart_params).sequences\n",
" indices = indices[:, :, 1:]\n",
"\n",
" images = p_get_images(indices, vqgan_params)\n",
" images = np.squeeze(np.asarray(images), 1)\n",
" for image in images:\n",
" all_images.append(custom_to_pil(image))\n",
" return all_images"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "831c715f",
"metadata": {},
"outputs": [],
"source": [
"def clip_top_k(prompt, images, k=8):\n",
" inputs = processor(text=prompt, images=images, return_tensors=\"np\", padding=True)\n",
" outputs = clip(**inputs)\n",
" logits = outputs.logits_per_text\n",
" scores = np.array(logits[0]).argsort()[-k:][::-1]\n",
" return [images[score] for score in scores]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "00605e13",
"metadata": {},
"outputs": [],
"source": [
"prompt = \"white snow covered mountain under blue sky during daytime\"\n",
"images = hallucinate(prompt)\n",
"selected = clip_top_k(prompt, images, k=8)\n",
"stack_reconstructions(selected)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cc745da2",
"metadata": {},
"outputs": [],
"source": [
"prompt = \"aerial view of beach at night\"\n",
"images = hallucinate(prompt)\n",
"selected = clip_top_k(prompt, images, k=8)\n",
"stack_reconstructions(selected)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c9cc0b1d",
"metadata": {},
"outputs": [],
"source": [
"prompt = \"an armchair in the shape of an avocado\"\n",
"images = hallucinate(prompt)\n",
"selected = clip_top_k(prompt, images, k=8)\n",
"stack_reconstructions(selected)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "574e9433",
"metadata": {},
"outputs": [],
"source": [
"prompt = \"young woman riding her bike into a forest\"\n",
"images = hallucinate(prompt)\n",
"selected = clip_top_k(prompt, images, k=8)\n",
"stack_reconstructions(selected)"
]
},
{
"cell_type": "markdown",
"id": "4762c91e",
"metadata": {},
"source": [
"`Forest` seems to dominate. Interesting cubist interpretation in the fourth image."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "af30608a",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.5"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|