deploy

Dockerfile

@@ -0,0 +1,14 @@
+# read the doc: https://huggingface.co/docs/hub/spaces-sdks-docker
+# you will also find guides on how best to write your Dockerfile
+
+FROM python:3.9
+
+WORKDIR /code
+
+COPY ./requirements.txt /code/requirements.txt
+
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+
+COPY . .
+
+CMD ["gunicorn", "-b", "0.0.0.0:7860", "main:app"]

lstm.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d64c990b25a04b2b15449f2280aab6cbafe3e8066099994fc9a88ccfc4ab2a6
+size 41351302

main.py

@@ -0,0 +1,51 @@
+from flask import Flask, request
+import torch
+import model
+import numpy as np
+from sentence_transformers import SentenceTransformer
+
+sentence_model = SentenceTransformer("all-MiniLM-L6-v2")
+
+embedding_dim = 384
+hidden_dim = 512
+num_layers = 1
+output_dim = 180
+num_epochs = 100
+learning_rate = 0.001
+
+lstm_model = model.LSTM(embedding_dim, hidden_dim, num_layers, output_dim)
+lstm_model.load_state_dict(torch.load('lstm.pt'))
+
+app = Flask(__name__)
+
+def GeneratePosesJSON(input):
+    with torch.no_grad():
+        processed_text = torch.tensor(sentence_model.encode(input), dtype=torch.float)
+        output_poses = lstm_model(processed_text.unsqueeze(0))
+
+        people = output_poses.cpu().detach().numpy().reshape(5, 18, 2).tolist()
+        newPeople = []
+        for person in people:
+            newPerson = []
+            for keypoints in person:
+                newPerson.append([keypoints[0], keypoints[1], 1])
+            newPeople.append(newPerson)
+
+        data = np.array(newPeople).reshape(5, 54).tolist()
+        formatted_data = []
+        for person in data:
+            formatted_data.append({ "pose_keypoints_2d": person })
+        return { 'people': data, 'animals': [], 'canvas_width': 900, 'canvas_height': 300 }
+    
+@app.route('/')
+def hello():
+    return "Hello, World!"
+
+@app.route('/generate')
+def generatePose():
+    text = request.args.get('text')
+    data = GeneratePosesJSON(text)
+    return data
+
+if __name__ == '__main__':
+    app.run()

model.py

@@ -0,0 +1,20 @@
+import torch.nn as nn
+
+class LSTM(nn.Module):
+    def __init__(self, embedding_dim, hidden_dim, num_layers, output_dim):
+        super(LSTM, self).__init__()
+        self.lstm1 = nn.LSTM(embedding_dim, hidden_dim, num_layers, batch_first=True)
+        self.lstm2 = nn.LSTM(hidden_dim, hidden_dim, num_layers, batch_first=True)
+        self.lstm3 = nn.LSTM(hidden_dim, hidden_dim, num_layers, batch_first=True)
+        self.lstm4 = nn.LSTM(hidden_dim, hidden_dim, num_layers, batch_first=True)
+        self.lstm5 = nn.LSTM(hidden_dim, hidden_dim, num_layers, batch_first=True)
+        self.o = nn.Linear(hidden_dim, output_dim)
+
+    def forward(self, embedding):
+        o_n1, (h_n1, c_n1) = self.lstm1(embedding)
+        o_n2, (h_n2, c_n2) = self.lstm2(o_n1, (h_n1, c_n1))
+        o_n3, (h_n3, c_n3) = self.lstm3(o_n2, (h_n2, c_n2))
+        o_n4, (h_n4, c_n4) = self.lstm4(o_n3, (h_n3, c_n3))
+        o_n5, (h_n5, c_n5) = self.lstm5(o_n4, (h_n4, c_n4))
+        output = self.o(o_n5)
+        return output

requirements.txt

@@ -0,0 +1,5 @@
+sentence_transformers
+numpy
+torch
+Flask
+Gunicorn