Spaces:
Runtime error
Runtime error
Fabrice-TIERCELIN
commited on
Commit
•
67efc35
1
Parent(s):
1b20dc8
Delete clipseg/Tables.ipynb
Browse files- clipseg/Tables.ipynb +0 -349
clipseg/Tables.ipynb
DELETED
@@ -1,349 +0,0 @@
|
|
1 |
-
{
|
2 |
-
"cells": [
|
3 |
-
{
|
4 |
-
"cell_type": "code",
|
5 |
-
"execution_count": null,
|
6 |
-
"metadata": {},
|
7 |
-
"outputs": [],
|
8 |
-
"source": [
|
9 |
-
"%load_ext autoreload\n",
|
10 |
-
"%autoreload 2\n",
|
11 |
-
"\n",
|
12 |
-
"import clip\n",
|
13 |
-
"from evaluation_utils import norm, denorm\n",
|
14 |
-
"from general_utils import *\n",
|
15 |
-
"from datasets.lvis_oneshot3 import LVIS_OneShot3, LVIS_OneShot"
|
16 |
-
]
|
17 |
-
},
|
18 |
-
{
|
19 |
-
"cell_type": "markdown",
|
20 |
-
"metadata": {},
|
21 |
-
"source": [
|
22 |
-
"# PhraseCut"
|
23 |
-
]
|
24 |
-
},
|
25 |
-
{
|
26 |
-
"cell_type": "code",
|
27 |
-
"execution_count": null,
|
28 |
-
"metadata": {},
|
29 |
-
"outputs": [],
|
30 |
-
"source": [
|
31 |
-
"pc = experiment('experiments/phrasecut.yaml', nums=':6').dataframe()"
|
32 |
-
]
|
33 |
-
},
|
34 |
-
{
|
35 |
-
"cell_type": "code",
|
36 |
-
"execution_count": null,
|
37 |
-
"metadata": {},
|
38 |
-
"outputs": [],
|
39 |
-
"source": [
|
40 |
-
"tab1 = pc[['name', 'pc_miou_best', 'pc_fgiou_best', 'pc_ap']]"
|
41 |
-
]
|
42 |
-
},
|
43 |
-
{
|
44 |
-
"cell_type": "code",
|
45 |
-
"execution_count": null,
|
46 |
-
"metadata": {},
|
47 |
-
"outputs": [],
|
48 |
-
"source": [
|
49 |
-
"cols = ['pc_miou_0.3', 'pc_fgiou_0.3', 'pc_ap']\n",
|
50 |
-
"tab1 = pc[['name'] + cols]\n",
|
51 |
-
"for k in cols:\n",
|
52 |
-
" tab1.loc[:, k] = (100 * tab1.loc[:, k]).round(1)\n",
|
53 |
-
"tab1.loc[:, 'name'] = ['CLIPSeg (PC+)', 'CLIPSeg (PC, $D=128$)', 'CLIPSeg (PC)', 'CLIP-Deconv', 'ViTSeg (PC+)', 'ViTSeg (PC)']\n",
|
54 |
-
"tab1.insert(1, 't', [0.3]*tab1.shape[0])\n",
|
55 |
-
"print(tab1.to_latex(header=False, index=False))"
|
56 |
-
]
|
57 |
-
},
|
58 |
-
{
|
59 |
-
"cell_type": "markdown",
|
60 |
-
"metadata": {},
|
61 |
-
"source": [
|
62 |
-
"For 0.1 threshold"
|
63 |
-
]
|
64 |
-
},
|
65 |
-
{
|
66 |
-
"cell_type": "code",
|
67 |
-
"execution_count": null,
|
68 |
-
"metadata": {},
|
69 |
-
"outputs": [],
|
70 |
-
"source": [
|
71 |
-
"cols = ['pc_miou_0.1', 'pc_fgiou_0.1', 'pc_ap']\n",
|
72 |
-
"tab1 = pc[['name'] + cols]\n",
|
73 |
-
"for k in cols:\n",
|
74 |
-
" tab1.loc[:, k] = (100 * tab1.loc[:, k]).round(1)\n",
|
75 |
-
"tab1.loc[:, 'name'] = ['CLIPSeg (PC+)', 'CLIPSeg (PC, $D=128$)', 'CLIPSeg (PC)', 'CLIP-Deconv', 'ViTSeg (PC+)', 'ViTSeg (PC)']\n",
|
76 |
-
"tab1.insert(1, 't', [0.1]*tab1.shape[0])\n",
|
77 |
-
"print(tab1.to_latex(header=False, index=False))"
|
78 |
-
]
|
79 |
-
},
|
80 |
-
{
|
81 |
-
"cell_type": "markdown",
|
82 |
-
"metadata": {},
|
83 |
-
"source": [
|
84 |
-
"# One-shot"
|
85 |
-
]
|
86 |
-
},
|
87 |
-
{
|
88 |
-
"cell_type": "markdown",
|
89 |
-
"metadata": {},
|
90 |
-
"source": [
|
91 |
-
"### Pascal"
|
92 |
-
]
|
93 |
-
},
|
94 |
-
{
|
95 |
-
"cell_type": "code",
|
96 |
-
"execution_count": null,
|
97 |
-
"metadata": {},
|
98 |
-
"outputs": [],
|
99 |
-
"source": [
|
100 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums=':19').dataframe()"
|
101 |
-
]
|
102 |
-
},
|
103 |
-
{
|
104 |
-
"cell_type": "code",
|
105 |
-
"execution_count": null,
|
106 |
-
"metadata": {},
|
107 |
-
"outputs": [],
|
108 |
-
"source": [
|
109 |
-
"pas[['name', 'pas_h2_miou_0.3', 'pas_h2_biniou_0.3', 'pas_h2_ap', 'pas_h2_fgiou_ct']]"
|
110 |
-
]
|
111 |
-
},
|
112 |
-
{
|
113 |
-
"cell_type": "code",
|
114 |
-
"execution_count": null,
|
115 |
-
"metadata": {},
|
116 |
-
"outputs": [],
|
117 |
-
"source": [
|
118 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums=':8').dataframe()\n",
|
119 |
-
"tab1 = pas[['pas_h2_miou_0.3', 'pas_h2_biniou_0.3', 'pas_h2_ap']]\n",
|
120 |
-
"print('CLIPSeg (PC+) & 0.3 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')\n",
|
121 |
-
"print('CLIPSeg (PC) & 0.3 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[4:8].mean(0).values), '\\\\\\\\')\n",
|
122 |
-
"\n",
|
123 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums='12:16').dataframe()\n",
|
124 |
-
"tab1 = pas[['pas_h2_miou_0.2', 'pas_h2_biniou_0.2', 'pas_h2_ap']]\n",
|
125 |
-
"print('CLIP-Deconv (PC+) & 0.2 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')\n",
|
126 |
-
"\n",
|
127 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums='16:20').dataframe()\n",
|
128 |
-
"tab1 = pas[['pas_t_miou_0.2', 'pas_t_biniou_0.2', 'pas_t_ap']]\n",
|
129 |
-
"print('ViTSeg (PC+) & 0.2 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')"
|
130 |
-
]
|
131 |
-
},
|
132 |
-
{
|
133 |
-
"cell_type": "markdown",
|
134 |
-
"metadata": {},
|
135 |
-
"source": [
|
136 |
-
"#### Pascal Zero-shot (in one-shot setting)\n",
|
137 |
-
"\n",
|
138 |
-
"Using the same setting as one-shot (hence different from the other zero-shot benchmark)"
|
139 |
-
]
|
140 |
-
},
|
141 |
-
{
|
142 |
-
"cell_type": "code",
|
143 |
-
"execution_count": null,
|
144 |
-
"metadata": {},
|
145 |
-
"outputs": [],
|
146 |
-
"source": [
|
147 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums=':8').dataframe()\n",
|
148 |
-
"tab1 = pas[['pas_t_miou_0.3', 'pas_t_biniou_0.3', 'pas_t_ap']]\n",
|
149 |
-
"print('CLIPSeg (PC+) & 0.3 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')\n",
|
150 |
-
"print('CLIPSeg (PC) & 0.3 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[4:8].mean(0).values), '\\\\\\\\')\n",
|
151 |
-
"\n",
|
152 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums='12:16').dataframe()\n",
|
153 |
-
"tab1 = pas[['pas_t_miou_0.3', 'pas_t_biniou_0.3', 'pas_t_ap']]\n",
|
154 |
-
"print('CLIP-Deconv (PC+) & 0.3 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')\n",
|
155 |
-
"\n",
|
156 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums='16:20').dataframe()\n",
|
157 |
-
"tab1 = pas[['pas_t_miou_0.2', 'pas_t_biniou_0.2', 'pas_t_ap']]\n",
|
158 |
-
"print('ViTSeg (PC+) & 0.2 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')"
|
159 |
-
]
|
160 |
-
},
|
161 |
-
{
|
162 |
-
"cell_type": "code",
|
163 |
-
"execution_count": null,
|
164 |
-
"metadata": {},
|
165 |
-
"outputs": [],
|
166 |
-
"source": [
|
167 |
-
"# without fixed thresholds...\n",
|
168 |
-
"\n",
|
169 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums=':8').dataframe()\n",
|
170 |
-
"tab1 = pas[['pas_t_best_miou', 'pas_t_best_biniou', 'pas_t_ap']]\n",
|
171 |
-
"print('CLIPSeg (PC+) & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')\n",
|
172 |
-
"print('CLIPSeg (PC) & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[4:8].mean(0).values), '\\\\\\\\')\n",
|
173 |
-
"\n",
|
174 |
-
"pas = experiment('experiments/pascal_1shot.yaml', nums='12:16').dataframe()\n",
|
175 |
-
"tab1 = pas[['pas_t_best_miou', 'pas_t_best_biniou', 'pas_t_ap']]\n",
|
176 |
-
"print('CLIP-Deconv (PC+) & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[0:4].mean(0).values), '\\\\\\\\')"
|
177 |
-
]
|
178 |
-
},
|
179 |
-
{
|
180 |
-
"cell_type": "markdown",
|
181 |
-
"metadata": {},
|
182 |
-
"source": [
|
183 |
-
"### COCO"
|
184 |
-
]
|
185 |
-
},
|
186 |
-
{
|
187 |
-
"cell_type": "code",
|
188 |
-
"execution_count": null,
|
189 |
-
"metadata": {},
|
190 |
-
"outputs": [],
|
191 |
-
"source": [
|
192 |
-
"coco = experiment('experiments/coco.yaml', nums=':29').dataframe()"
|
193 |
-
]
|
194 |
-
},
|
195 |
-
{
|
196 |
-
"cell_type": "code",
|
197 |
-
"execution_count": null,
|
198 |
-
"metadata": {},
|
199 |
-
"outputs": [],
|
200 |
-
"source": [
|
201 |
-
"tab1 = coco[['coco_h2_miou_0.1', 'coco_h2_biniou_0.1', 'coco_h2_ap']]\n",
|
202 |
-
"tab2 = coco[['coco_h2_miou_0.2', 'coco_h2_biniou_0.2', 'coco_h2_ap']]\n",
|
203 |
-
"tab3 = coco[['coco_h2_miou_best', 'coco_h2_biniou_best', 'coco_h2_ap']]\n",
|
204 |
-
"print('CLIPSeg (COCO) & 0.1 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[:4].mean(0).values), '\\\\\\\\')\n",
|
205 |
-
"print('CLIPSeg (COCO+N) & 0.1 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[4:8].mean(0).values), '\\\\\\\\')\n",
|
206 |
-
"print('CLIP-Deconv (COCO+N) & 0.1 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[12:16].mean(0).values), '\\\\\\\\')\n",
|
207 |
-
"print('ViTSeg (COCO) & 0.1 & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[8:12].mean(0).values), '\\\\\\\\')"
|
208 |
-
]
|
209 |
-
},
|
210 |
-
{
|
211 |
-
"cell_type": "markdown",
|
212 |
-
"metadata": {},
|
213 |
-
"source": [
|
214 |
-
"# Zero-shot"
|
215 |
-
]
|
216 |
-
},
|
217 |
-
{
|
218 |
-
"cell_type": "code",
|
219 |
-
"execution_count": null,
|
220 |
-
"metadata": {},
|
221 |
-
"outputs": [],
|
222 |
-
"source": [
|
223 |
-
"zs = experiment('experiments/pascal_0shot.yaml', nums=':11').dataframe()"
|
224 |
-
]
|
225 |
-
},
|
226 |
-
{
|
227 |
-
"cell_type": "code",
|
228 |
-
"execution_count": null,
|
229 |
-
"metadata": {},
|
230 |
-
"outputs": [],
|
231 |
-
"source": [
|
232 |
-
"\n",
|
233 |
-
"tab1 = zs[['pas_zs_seen', 'pas_zs_unseen']]\n",
|
234 |
-
"print('CLIPSeg (PC+) & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[8:9].values[0].tolist() + tab1[10:11].values[0].tolist()), '\\\\\\\\')\n",
|
235 |
-
"print('CLIP-Deconv & CLIP & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[2:3].values[0].tolist() + tab1[3:4].values[0].tolist()), '\\\\\\\\')\n",
|
236 |
-
"print('ViTSeg & ImageNet-1K & ' + ' & '.join(f'{x*100:.1f}' for x in tab1[4:5].values[0].tolist() + tab1[5:6].values[0].tolist()), '\\\\\\\\')"
|
237 |
-
]
|
238 |
-
},
|
239 |
-
{
|
240 |
-
"cell_type": "markdown",
|
241 |
-
"metadata": {},
|
242 |
-
"source": [
|
243 |
-
"# Ablation"
|
244 |
-
]
|
245 |
-
},
|
246 |
-
{
|
247 |
-
"cell_type": "code",
|
248 |
-
"execution_count": null,
|
249 |
-
"metadata": {},
|
250 |
-
"outputs": [],
|
251 |
-
"source": [
|
252 |
-
"ablation = experiment('experiments/ablation.yaml', nums=':8').dataframe()"
|
253 |
-
]
|
254 |
-
},
|
255 |
-
{
|
256 |
-
"cell_type": "code",
|
257 |
-
"execution_count": null,
|
258 |
-
"metadata": {},
|
259 |
-
"outputs": [],
|
260 |
-
"source": [
|
261 |
-
"tab1 = ablation[['name', 'pc_miou_best', 'pc_ap', 'pc-vis_miou_best', 'pc-vis_ap']]\n",
|
262 |
-
"for k in ['pc_miou_best', 'pc_ap', 'pc-vis_miou_best', 'pc-vis_ap']:\n",
|
263 |
-
" tab1.loc[:, k] = (100 * tab1.loc[:, k]).round(1)\n",
|
264 |
-
"tab1.loc[:, 'name'] = ['CLIPSeg', 'no CLIP pre-training', 'no-negatives', '50% negatives', 'no visual', '$D=16$', 'only layer 3', 'highlight mask']"
|
265 |
-
]
|
266 |
-
},
|
267 |
-
{
|
268 |
-
"cell_type": "code",
|
269 |
-
"execution_count": null,
|
270 |
-
"metadata": {},
|
271 |
-
"outputs": [],
|
272 |
-
"source": [
|
273 |
-
"print(tab1.loc[[0,1,4,5,6,7],:].to_latex(header=False, index=False))"
|
274 |
-
]
|
275 |
-
},
|
276 |
-
{
|
277 |
-
"cell_type": "code",
|
278 |
-
"execution_count": null,
|
279 |
-
"metadata": {},
|
280 |
-
"outputs": [],
|
281 |
-
"source": [
|
282 |
-
"print(tab1.loc[[0,1,4,5,6,7],:].to_latex(header=False, index=False))"
|
283 |
-
]
|
284 |
-
},
|
285 |
-
{
|
286 |
-
"cell_type": "markdown",
|
287 |
-
"metadata": {},
|
288 |
-
"source": [
|
289 |
-
"# Generalization"
|
290 |
-
]
|
291 |
-
},
|
292 |
-
{
|
293 |
-
"cell_type": "code",
|
294 |
-
"execution_count": null,
|
295 |
-
"metadata": {},
|
296 |
-
"outputs": [],
|
297 |
-
"source": [
|
298 |
-
"generalization = experiment('experiments/generalize.yaml').dataframe()"
|
299 |
-
]
|
300 |
-
},
|
301 |
-
{
|
302 |
-
"cell_type": "code",
|
303 |
-
"execution_count": null,
|
304 |
-
"metadata": {},
|
305 |
-
"outputs": [],
|
306 |
-
"source": [
|
307 |
-
"gen = generalization[['aff_best_fgiou', 'aff_ap', 'ability_best_fgiou', 'ability_ap', 'part_best_fgiou', 'part_ap']].values"
|
308 |
-
]
|
309 |
-
},
|
310 |
-
{
|
311 |
-
"cell_type": "code",
|
312 |
-
"execution_count": null,
|
313 |
-
"metadata": {},
|
314 |
-
"outputs": [],
|
315 |
-
"source": [
|
316 |
-
"print(\n",
|
317 |
-
" 'CLIPSeg (PC+) & ' + ' & '.join(f'{x*100:.1f}' for x in gen[1]) + ' \\\\\\\\ \\n' + \\\n",
|
318 |
-
" 'CLIPSeg (LVIS) & ' + ' & '.join(f'{x*100:.1f}' for x in gen[0]) + ' \\\\\\\\ \\n' + \\\n",
|
319 |
-
" 'CLIP-Deconv & ' + ' & '.join(f'{x*100:.1f}' for x in gen[2]) + ' \\\\\\\\ \\n' + \\\n",
|
320 |
-
" 'VITSeg & ' + ' & '.join(f'{x*100:.1f}' for x in gen[3]) + ' \\\\\\\\'\n",
|
321 |
-
")"
|
322 |
-
]
|
323 |
-
}
|
324 |
-
],
|
325 |
-
"metadata": {
|
326 |
-
"interpreter": {
|
327 |
-
"hash": "800ed241f7db2bd3aa6942aa3be6809cdb30ee6b0a9e773dfecfa9fef1f4c586"
|
328 |
-
},
|
329 |
-
"kernelspec": {
|
330 |
-
"display_name": "env2",
|
331 |
-
"language": "python",
|
332 |
-
"name": "env2"
|
333 |
-
},
|
334 |
-
"language_info": {
|
335 |
-
"codemirror_mode": {
|
336 |
-
"name": "ipython",
|
337 |
-
"version": 3
|
338 |
-
},
|
339 |
-
"file_extension": ".py",
|
340 |
-
"mimetype": "text/x-python",
|
341 |
-
"name": "python",
|
342 |
-
"nbconvert_exporter": "python",
|
343 |
-
"pygments_lexer": "ipython3",
|
344 |
-
"version": "3.8.8"
|
345 |
-
}
|
346 |
-
},
|
347 |
-
"nbformat": 4,
|
348 |
-
"nbformat_minor": 4
|
349 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|