Add application and pretraied model

app.py

@@ -0,0 +1,208 @@
+from pathlib import Path
+import numpy as np
+import random
+import re
+import textwrap
+
+from shapely.geometry.polygon import Polygon
+import aggdraw
+from PIL import Image, ImageDraw, ImageOps, ImageFilter, ImageFont, ImageColor
+
+import gradio as gr
+
+from transformers import AutoTokenizer, AutoConfig, AutoModelForCausalLM
+
+finetuned = AutoModelForCausalLM.from_pretrained('model')
+tokenizer = AutoTokenizer.from_pretrained('gpt2')
+
+# Utility functions
+
+housegan_labels = {"living_room": 1, "kitchen": 2, "bedroom": 3, "bathroom": 4, "missing": 5, "closet": 6, 
+                         "balcony": 7, "corridor": 8, "dining_room": 9, "laundry_room": 10}
+
+housegan_colors = [[0, 0, 0], [197, 203, 159], [169, 89, 142], [0, 132, 66], [190, 0, 198], [255, 255, 255], 
+                   [6, 53, 17], [2, 54, 192], [132, 151, 246], [197, 203, 159], [6, 53, 17],]
+
+regex = re.compile(".*?\((.*?)\)")
+
+def draw_polygons(polygons, colors, im_size=(256, 256), b_color="white", fpath=None):
+
+    image = Image.new("RGB", im_size, color="white")
+    draw = aggdraw.Draw(image)
+
+    for poly, color, in zip(polygons, colors):
+        #get initial polygon coordinates
+        xy = poly.exterior.xy
+        coords = np.dstack((xy[1], xy[0])).flatten()
+        # draw it on canvas, with the appropriate colors
+        brush = aggdraw.Brush((0, 0, 0), opacity=255)
+        draw.polygon(coords, brush)
+        
+        
+        #get inner polygon coordinates
+        small_poly = poly.buffer(-1, resolution=32, cap_style=2, join_style=2, mitre_limit=5.0)
+        if small_poly.geom_type == 'MultiPolygon':
+            mycoordslist = [list(x.exterior.coords) for x in small_poly]
+            for coord in mycoordslist:
+                coords = np.dstack((np.array(coord)[:,1], np.array(coord)[:, 0])).flatten()
+                brush2 = aggdraw.Brush((0, 0, 0), opacity=255)
+                draw.polygon(coords, brush2)
+        elif poly.geom_type == 'Polygon':
+            #get inner polygon coordinates
+            xy2 = small_poly.exterior.xy
+            coords2 = np.dstack((xy2[1], xy2[0])).flatten()
+            # draw it on canvas, with the appropriate colors
+            brush2 = aggdraw.Brush((color[0], color[1], color[2]), opacity=255)
+            draw.polygon(coords2, brush2)
+
+    image = Image.frombytes("RGB", (256,256), draw.tobytes()).transpose(Image.FLIP_TOP_BOTTOM)
+
+    if(fpath):
+        image.save(fpath, quality=100, subsampling=0)
+
+    return draw, image
+
+def prompt_to_layout(user_prompt, fpath=None):
+    
+    model_prompt = '[User prompt] {} [Layout]'.format(user_prompt)
+    input_ids = tokenizer(model_prompt, return_tensors='pt')
+    output = finetuned.generate(**input_ids, do_sample=True, top_p=0.94, top_k=100, max_length=300)
+    output = tokenizer.batch_decode(output, skip_special_tokens=True)
+    output = output[0].split('[Layout]')[1].split(', ')
+    spaces = [txt.split(':')[0] for txt in output]
+
+    coordinates = [txt.split(':')[1] for txt in output]
+    coordinates = [re.findall(regex, coord) for coord in coordinates]
+    
+    polygons = []
+    for coord in coordinates:
+        polygons.append([point.split(',') for point in coord])
+        
+    geom = []
+    for poly in polygons:
+        geom.append(Polygon(np.array(poly, dtype=int)))
+        
+    colors = [housegan_colors[housegan_labels[space]] for space in spaces]
+    
+    _, im = draw_polygons(geom, colors, fpath=fpath)
+    
+    legend = Image.open(r"C:\\Users\\user\\Desktop\\legend3.png")
+    
+    im = np.array(im)
+    im[:40, :] = np.array(legend)
+    im = Image.fromarray(im)
+    
+    return im, output
+    
+def mut_txt2layout(mut_output):
+    output = mut_output[0].split('[Layout]')[1].split(', ')
+    spaces = [txt.split(':')[0].strip(' ') for txt in output]
+    coordinates = [txt.split(':')[1] for txt in output]
+    coordinates = [re.findall(regex, coord) for coord in coordinates]
+
+    polygons = []
+    for coord in coordinates:
+        polygons.append([point.split(',') for point in coord])
+
+    geom = []
+    for poly in polygons:
+        geom.append(Polygon(np.array(poly, dtype=int)))
+
+    colors = [housegan_colors[housegan_labels[space]] for space in spaces]
+    _, im = draw_polygons(geom, colors, fpath=None)
+
+    legend = Image.open(r"C:\\Users\\user\\Desktop\\legend3.png")
+
+    im = np.array(im)
+    im[:40, :] = np.array(legend)
+    im = Image.fromarray(im)
+    
+    return im
+    
+def prompt_with_mutation(user_prompt, fpath=None):
+    
+    #Create initial layout based on prompt
+    im, output = prompt_to_layout(user_prompt)
+        
+    #Create mutated layout based on initial
+    cut_off = np.random.randint(1, 3, size=1)[0]
+    cut_off = min(cut_off, len(output)-1)
+    new_prompt = model_prompt + ', '.join(output[:cut_off]) + ', '
+    input_ids = tokenizer(new_prompt, return_tensors='pt')
+    mut_output = finetuned.generate(**input_ids, do_sample=True, top_p=0.94, top_k=100, max_length=200)
+    mut_output = tokenizer.batch_decode(mut_output, skip_special_tokens=True)
+    mut_im = mut_txt2layout(mut_output)
+    
+    combined = merge_images(im, mut_im)
+    
+    return im, mut_im
+    
+# Gradio App
+
+def gen_and_mutate(user_prompt, mutate=False):    
+    if(mutate):
+        im, mut_im = None, None
+        while (mut_im is None):
+            im, mut_im = prompt_with_mutation(user_prompt)
+    else:
+        mut_im=Image.open(r"C:\\Users\\user\\Desktop\\empty.png")
+        im, _ = prompt_to_layout(user_prompt)
+
+    return im, mut_im
+    
+checkbox =  gr.inputs.Checkbox(label='Mutate')
+textbox = gr.inputs.Textbox(placeholder='Enter a prompt describing a layout, see below for instructions')
+
+generated = gr.outputs.Image(label='Generated Layout')
+mutated = gr.outputs.Image(label='Mutated Layout')
+
+iface = gr.Interface(fn=gen_and_mutate, inputs=[textbox, checkbox], outputs=[generated, mutated],
+                      thumbnail=r"E:\Datasets\MyFloorplans\text2text\thumbnail_gradio.PNG",
+                      description='Demo of Semantic Generation of Residential Layouts \n',
+                      article='''<div>
+    <p> This app allows users the use of natural language prompts for appartment layout generation, using a variety of semantic information:</p>
+     <ul>
+      <li> <strong>typology</strong>: "a bedroom with two bedrooms and two bathrooms"</li>
+      <li> <strong>enumeration</strong>: "a house with five rooms"</li>
+      <li> <strong>adjacency</strong>: "the kitchen is adjacent to a bedroom", "the living room is not adjacent to the bathroom"</li>
+      <li> <strong>location</strong>: "a house with a bedroom in the north east side"</li>
+    </ul>
+    <p>You can also create a mutation of the generated layout by enabling the 'Mutate' option.</p>
+    <p> Made by: <a href='https://www.linkedin.com/in/theodorosgalanos/'>Theodoros </a> <a href='https://twitter.com/TheodoreGalanos'> Galanos</a> and <a href='https://twitter.com/tylerlastovich'>Tyler Lastovich</a> </p>
+</div>''')
+
+iface.launch()
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

model/config.json

@@ -0,0 +1,53 @@
+{
+  "_name_or_path": "EleutherAI/gpt-neo-125M",
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPTNeoForCausalLM"
+  ],
+  "attention_dropout": 0,
+  "attention_layers": [
+    "global",
+    "local",
+    "global",
+    "local",
+    "global",
+    "local",
+    "global",
+    "local",
+    "global",
+    "local",
+    "global",
+    "local"
+  ],
+  "attention_types": [
+    [
+      [
+        "global",
+        "local"
+      ],
+      6
+    ]
+  ],
+  "bos_token_id": 50256,
+  "embed_dropout": 0,
+  "eos_token_id": 50256,
+  "gradient_checkpointing": false,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": null,
+  "layer_norm_epsilon": 1e-05,
+  "max_position_embeddings": 2048,
+  "model_type": "gpt_neo",
+  "num_heads": 12,
+  "num_layers": 12,
+  "resid_dropout": 0,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "transformers_version": "4.8.1",
+  "use_cache": true,
+  "vocab_size": 50257,
+  "window_size": 256
+}

model/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df35819ca3ee7d6667b734dcd2a92c2ab539a72b399f8fb76cdbd89ad8a57df5
+size 526021937

model/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6081874f960bdd9087fced5865fc0df1593c2ffcde9c679c276725789565994
+size 15691

model/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4175248537ab98940ac983894a7f290b125e0bdf7b03a2c00c7c4855d080b5f
+size 623

model/trainer_state.json

@@ -0,0 +1,1196 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0,
+  "global_step": 97022,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.25e-06,
+      "loss": 0.3381,
+      "step": 500
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 2.5e-06,
+      "loss": 0.1717,
+      "step": 1000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.7500000000000005e-06,
+      "loss": 0.1167,
+      "step": 1500
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 5e-06,
+      "loss": 0.0962,
+      "step": 2000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.9896073662987675e-06,
+      "loss": 0.0848,
+      "step": 2500
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.979214732597535e-06,
+      "loss": 0.0796,
+      "step": 3000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.968822098896302e-06,
+      "loss": 0.0758,
+      "step": 3500
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.95842946519507e-06,
+      "loss": 0.0732,
+      "step": 4000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.948036831493837e-06,
+      "loss": 0.071,
+      "step": 4500
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.937644197792605e-06,
+      "loss": 0.0697,
+      "step": 5000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.927251564091372e-06,
+      "loss": 0.0681,
+      "step": 5500
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.91685893039014e-06,
+      "loss": 0.0678,
+      "step": 6000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.906466296688907e-06,
+      "loss": 0.0656,
+      "step": 6500
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.896073662987675e-06,
+      "loss": 0.0651,
+      "step": 7000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.885681029286442e-06,
+      "loss": 0.0644,
+      "step": 7500
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.87528839558521e-06,
+      "loss": 0.0638,
+      "step": 8000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.864895761883977e-06,
+      "loss": 0.0627,
+      "step": 8500
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.854503128182744e-06,
+      "loss": 0.0622,
+      "step": 9000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.844110494481512e-06,
+      "loss": 0.0615,
+      "step": 9500
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.833717860780279e-06,
+      "loss": 0.061,
+      "step": 10000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.823325227079046e-06,
+      "loss": 0.0607,
+      "step": 10500
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.812932593377814e-06,
+      "loss": 0.0603,
+      "step": 11000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.802539959676581e-06,
+      "loss": 0.0597,
+      "step": 11500
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.792147325975349e-06,
+      "loss": 0.0593,
+      "step": 12000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.781754692274117e-06,
+      "loss": 0.0588,
+      "step": 12500
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.771362058572884e-06,
+      "loss": 0.0587,
+      "step": 13000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.760969424871652e-06,
+      "loss": 0.058,
+      "step": 13500
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.750576791170419e-06,
+      "loss": 0.0578,
+      "step": 14000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.740184157469187e-06,
+      "loss": 0.0574,
+      "step": 14500
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.729791523767954e-06,
+      "loss": 0.0572,
+      "step": 15000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.719398890066721e-06,
+      "loss": 0.0566,
+      "step": 15500
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 4.709006256365489e-06,
+      "loss": 0.0566,
+      "step": 16000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 4.698613622664256e-06,
+      "loss": 0.0562,
+      "step": 16500
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 4.688220988963023e-06,
+      "loss": 0.0555,
+      "step": 17000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 4.677828355261791e-06,
+      "loss": 0.0554,
+      "step": 17500
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 4.667435721560558e-06,
+      "loss": 0.0559,
+      "step": 18000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 4.657043087859326e-06,
+      "loss": 0.0553,
+      "step": 18500
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 4.646650454158093e-06,
+      "loss": 0.0548,
+      "step": 19000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 4.636257820456861e-06,
+      "loss": 0.0548,
+      "step": 19500
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 4.6258651867556285e-06,
+      "loss": 0.0544,
+      "step": 20000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 4.6154725530543956e-06,
+      "loss": 0.0542,
+      "step": 20500
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 4.6050799193531635e-06,
+      "loss": 0.0542,
+      "step": 21000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 4.5946872856519305e-06,
+      "loss": 0.0539,
+      "step": 21500
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 4.5842946519506976e-06,
+      "loss": 0.0538,
+      "step": 22000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 4.5739020182494655e-06,
+      "loss": 0.0535,
+      "step": 22500
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 4.5635093845482325e-06,
+      "loss": 0.0534,
+      "step": 23000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 4.5531167508469996e-06,
+      "loss": 0.0535,
+      "step": 23500
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 4.5427241171457675e-06,
+      "loss": 0.0533,
+      "step": 24000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.5323314834445345e-06,
+      "loss": 0.053,
+      "step": 24500
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.521938849743302e-06,
+      "loss": 0.0529,
+      "step": 25000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.5115462160420694e-06,
+      "loss": 0.0527,
+      "step": 25500
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.501153582340837e-06,
+      "loss": 0.0524,
+      "step": 26000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.490760948639604e-06,
+      "loss": 0.052,
+      "step": 26500
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.480368314938372e-06,
+      "loss": 0.0522,
+      "step": 27000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.469975681237139e-06,
+      "loss": 0.0522,
+      "step": 27500
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.459583047535907e-06,
+      "loss": 0.0519,
+      "step": 28000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.449190413834674e-06,
+      "loss": 0.0521,
+      "step": 28500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.438797780133442e-06,
+      "loss": 0.0518,
+      "step": 29000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 4.428405146432209e-06,
+      "loss": 0.0515,
+      "step": 29500
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 4.418012512730976e-06,
+      "loss": 0.0514,
+      "step": 30000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 4.407619879029744e-06,
+      "loss": 0.0511,
+      "step": 30500
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 4.397227245328511e-06,
+      "loss": 0.0512,
+      "step": 31000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 4.386834611627279e-06,
+      "loss": 0.0509,
+      "step": 31500
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 4.376441977926046e-06,
+      "loss": 0.0513,
+      "step": 32000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 4.366049344224814e-06,
+      "loss": 0.0507,
+      "step": 32500
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 4.355656710523581e-06,
+      "loss": 0.0506,
+      "step": 33000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 4.345264076822349e-06,
+      "loss": 0.0503,
+      "step": 33500
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 4.334871443121116e-06,
+      "loss": 0.0501,
+      "step": 34000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 4.324478809419884e-06,
+      "loss": 0.0502,
+      "step": 34500
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 4.314086175718651e-06,
+      "loss": 0.0505,
+      "step": 35000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 4.303693542017419e-06,
+      "loss": 0.0505,
+      "step": 35500
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 4.293300908316186e-06,
+      "loss": 0.0501,
+      "step": 36000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 4.282908274614953e-06,
+      "loss": 0.05,
+      "step": 36500
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 4.272515640913721e-06,
+      "loss": 0.05,
+      "step": 37000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 4.262123007212488e-06,
+      "loss": 0.0498,
+      "step": 37500
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 4.251730373511256e-06,
+      "loss": 0.0498,
+      "step": 38000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 4.241337739810023e-06,
+      "loss": 0.0496,
+      "step": 38500
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 4.23094510610879e-06,
+      "loss": 0.0496,
+      "step": 39000
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.220552472407558e-06,
+      "loss": 0.0493,
+      "step": 39500
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 4.210159838706326e-06,
+      "loss": 0.0495,
+      "step": 40000
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 4.199767205005093e-06,
+      "loss": 0.049,
+      "step": 40500
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 4.189374571303861e-06,
+      "loss": 0.0493,
+      "step": 41000
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 4.178981937602628e-06,
+      "loss": 0.0491,
+      "step": 41500
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 4.1685893039013956e-06,
+      "loss": 0.0492,
+      "step": 42000
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 4.158196670200163e-06,
+      "loss": 0.0492,
+      "step": 42500
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 4.14780403649893e-06,
+      "loss": 0.0487,
+      "step": 43000
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.1374114027976976e-06,
+      "loss": 0.049,
+      "step": 43500
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.127018769096465e-06,
+      "loss": 0.0487,
+      "step": 44000
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.116626135395232e-06,
+      "loss": 0.0489,
+      "step": 44500
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 4.1062335016939995e-06,
+      "loss": 0.0485,
+      "step": 45000
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.095840867992767e-06,
+      "loss": 0.0486,
+      "step": 45500
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 4.0854482342915345e-06,
+      "loss": 0.0485,
+      "step": 46000
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 4.0750556005903015e-06,
+      "loss": 0.0487,
+      "step": 46500
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 4.0646629668890694e-06,
+      "loss": 0.0482,
+      "step": 47000
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 4.054270333187837e-06,
+      "loss": 0.0482,
+      "step": 47500
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.043877699486604e-06,
+      "loss": 0.0481,
+      "step": 48000
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.033485065785372e-06,
+      "loss": 0.0481,
+      "step": 48500
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.04874278977513313,
+      "eval_runtime": 1479.0093,
+      "eval_samples_per_second": 13.382,
+      "eval_steps_per_second": 0.836,
+      "step": 48511
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 4.023092432084139e-06,
+      "loss": 0.0479,
+      "step": 49000
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 4.012699798382906e-06,
+      "loss": 0.0478,
+      "step": 49500
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 4.002307164681674e-06,
+      "loss": 0.048,
+      "step": 50000
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 3.991914530980441e-06,
+      "loss": 0.0477,
+      "step": 50500
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 3.981521897279208e-06,
+      "loss": 0.0478,
+      "step": 51000
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 3.971129263577976e-06,
+      "loss": 0.0476,
+      "step": 51500
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 3.960736629876743e-06,
+      "loss": 0.0475,
+      "step": 52000
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 3.950343996175511e-06,
+      "loss": 0.0473,
+      "step": 52500
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 3.939951362474278e-06,
+      "loss": 0.0477,
+      "step": 53000
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 3.929558728773046e-06,
+      "loss": 0.0472,
+      "step": 53500
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 3.919166095071813e-06,
+      "loss": 0.0471,
+      "step": 54000
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 3.908773461370581e-06,
+      "loss": 0.0472,
+      "step": 54500
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 3.898380827669349e-06,
+      "loss": 0.0472,
+      "step": 55000
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 3.887988193968116e-06,
+      "loss": 0.0471,
+      "step": 55500
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 3.877595560266883e-06,
+      "loss": 0.0472,
+      "step": 56000
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 3.867202926565651e-06,
+      "loss": 0.0471,
+      "step": 56500
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 3.856810292864418e-06,
+      "loss": 0.047,
+      "step": 57000
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 3.846417659163185e-06,
+      "loss": 0.0472,
+      "step": 57500
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 3.836025025461953e-06,
+      "loss": 0.0471,
+      "step": 58000
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 3.82563239176072e-06,
+      "loss": 0.0468,
+      "step": 58500
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 3.815239758059488e-06,
+      "loss": 0.047,
+      "step": 59000
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 3.8048471243582554e-06,
+      "loss": 0.0467,
+      "step": 59500
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 3.7944544906570224e-06,
+      "loss": 0.0465,
+      "step": 60000
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 3.7840618569557903e-06,
+      "loss": 0.0467,
+      "step": 60500
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 3.7736692232545574e-06,
+      "loss": 0.0465,
+      "step": 61000
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 3.763276589553325e-06,
+      "loss": 0.0467,
+      "step": 61500
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 3.7528839558520923e-06,
+      "loss": 0.0468,
+      "step": 62000
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 3.7424913221508598e-06,
+      "loss": 0.0469,
+      "step": 62500
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 3.7320986884496272e-06,
+      "loss": 0.0465,
+      "step": 63000
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 3.7217060547483947e-06,
+      "loss": 0.0462,
+      "step": 63500
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 3.7113134210471618e-06,
+      "loss": 0.0467,
+      "step": 64000
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 3.7009207873459297e-06,
+      "loss": 0.0465,
+      "step": 64500
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 3.6905281536446967e-06,
+      "loss": 0.0464,
+      "step": 65000
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 3.6801355199434646e-06,
+      "loss": 0.0465,
+      "step": 65500
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 3.6697428862422316e-06,
+      "loss": 0.0465,
+      "step": 66000
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 3.659350252540999e-06,
+      "loss": 0.0463,
+      "step": 66500
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 3.648957618839767e-06,
+      "loss": 0.0461,
+      "step": 67000
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 3.638564985138534e-06,
+      "loss": 0.0461,
+      "step": 67500
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 3.628172351437301e-06,
+      "loss": 0.0462,
+      "step": 68000
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 3.617779717736069e-06,
+      "loss": 0.0459,
+      "step": 68500
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 3.6073870840348365e-06,
+      "loss": 0.0462,
+      "step": 69000
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 3.596994450333604e-06,
+      "loss": 0.0459,
+      "step": 69500
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 3.5866018166323714e-06,
+      "loss": 0.0459,
+      "step": 70000
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 3.5762091829311385e-06,
+      "loss": 0.0457,
+      "step": 70500
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 3.5658165492299064e-06,
+      "loss": 0.0457,
+      "step": 71000
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 3.5554239155286734e-06,
+      "loss": 0.0456,
+      "step": 71500
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 3.5450312818274413e-06,
+      "loss": 0.0457,
+      "step": 72000
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 3.5346386481262083e-06,
+      "loss": 0.0461,
+      "step": 72500
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 3.524246014424976e-06,
+      "loss": 0.046,
+      "step": 73000
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 3.5138533807237433e-06,
+      "loss": 0.0456,
+      "step": 73500
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 3.5034607470225108e-06,
+      "loss": 0.0456,
+      "step": 74000
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 3.493068113321278e-06,
+      "loss": 0.0458,
+      "step": 74500
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 3.4826754796200457e-06,
+      "loss": 0.0455,
+      "step": 75000
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 3.4722828459188128e-06,
+      "loss": 0.0458,
+      "step": 75500
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 3.4618902122175806e-06,
+      "loss": 0.0455,
+      "step": 76000
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 3.451497578516348e-06,
+      "loss": 0.0459,
+      "step": 76500
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 3.441104944815115e-06,
+      "loss": 0.0455,
+      "step": 77000
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 3.430712311113883e-06,
+      "loss": 0.0457,
+      "step": 77500
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 3.42031967741265e-06,
+      "loss": 0.0456,
+      "step": 78000
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 3.409927043711417e-06,
+      "loss": 0.0453,
+      "step": 78500
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 3.399534410010185e-06,
+      "loss": 0.0454,
+      "step": 79000
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 3.3891417763089525e-06,
+      "loss": 0.0453,
+      "step": 79500
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 3.37874914260772e-06,
+      "loss": 0.0455,
+      "step": 80000
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 3.3683565089064875e-06,
+      "loss": 0.0453,
+      "step": 80500
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 3.3579638752052545e-06,
+      "loss": 0.0455,
+      "step": 81000
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 3.3475712415040224e-06,
+      "loss": 0.0451,
+      "step": 81500
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 3.3371786078027895e-06,
+      "loss": 0.0448,
+      "step": 82000
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 3.3267859741015574e-06,
+      "loss": 0.0453,
+      "step": 82500
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 3.3163933404003244e-06,
+      "loss": 0.045,
+      "step": 83000
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 3.306000706699092e-06,
+      "loss": 0.045,
+      "step": 83500
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 3.2956080729978598e-06,
+      "loss": 0.0453,
+      "step": 84000
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 3.285215439296627e-06,
+      "loss": 0.045,
+      "step": 84500
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 3.274822805595394e-06,
+      "loss": 0.0454,
+      "step": 85000
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 3.2644301718941618e-06,
+      "loss": 0.045,
+      "step": 85500
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 3.254037538192929e-06,
+      "loss": 0.0449,
+      "step": 86000
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 3.2436449044916967e-06,
+      "loss": 0.045,
+      "step": 86500
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 3.233252270790464e-06,
+      "loss": 0.0448,
+      "step": 87000
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 3.2228596370892312e-06,
+      "loss": 0.0449,
+      "step": 87500
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 3.212467003387999e-06,
+      "loss": 0.0446,
+      "step": 88000
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 3.202074369686766e-06,
+      "loss": 0.045,
+      "step": 88500
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 3.191681735985534e-06,
+      "loss": 0.0449,
+      "step": 89000
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 3.181289102284301e-06,
+      "loss": 0.0446,
+      "step": 89500
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 3.1708964685830686e-06,
+      "loss": 0.0445,
+      "step": 90000
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 3.160503834881836e-06,
+      "loss": 0.0448,
+      "step": 90500
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 3.1501112011806035e-06,
+      "loss": 0.0445,
+      "step": 91000
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 3.1397185674793706e-06,
+      "loss": 0.0447,
+      "step": 91500
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 3.1293259337781385e-06,
+      "loss": 0.0446,
+      "step": 92000
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 3.1189333000769055e-06,
+      "loss": 0.0449,
+      "step": 92500
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 3.1085406663756734e-06,
+      "loss": 0.0446,
+      "step": 93000
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 3.0981480326744404e-06,
+      "loss": 0.0442,
+      "step": 93500
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 3.087755398973208e-06,
+      "loss": 0.0445,
+      "step": 94000
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 3.077362765271976e-06,
+      "loss": 0.0444,
+      "step": 94500
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 3.066970131570743e-06,
+      "loss": 0.0444,
+      "step": 95000
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 3.05657749786951e-06,
+      "loss": 0.0444,
+      "step": 95500
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 3.046184864168278e-06,
+      "loss": 0.0444,
+      "step": 96000
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 3.0357922304670453e-06,
+      "loss": 0.0446,
+      "step": 96500
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 3.0253995967658127e-06,
+      "loss": 0.0443,
+      "step": 97000
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 0.04501689225435257,
+      "eval_runtime": 1478.7113,
+      "eval_samples_per_second": 13.385,
+      "eval_steps_per_second": 0.837,
+      "step": 97022
+    }
+  ],
+  "max_steps": 242555,
+  "num_train_epochs": 5,
+  "total_flos": 1.1940942833642373e+18,
+  "trial_name": null,
+  "trial_params": null
+}

model/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51ee1474c90d834df17eec7e46d87cc82f364065cbe182dd8829d6f16c838280
+size 2735

requirements.txt

@@ -0,0 +1,8 @@
+numpy>=1.16.4
+textwrap3==0.9.2
+Pillow==8.0.1
+transformers==4.8.0
+gradio>=2.0.7
+shapely==1.7.1
+aggdraw==1.3.12
+torch==1.7.1