Menglinucas / Floorplan-recognition

Floorplan-recognition

---

1. Distinguishing floor plan image

Structure:

(1) (input)64x64x3 ---> conv + pool

(2) 32x32x32 ---> conv + pool

(3) 16x16x64 ---> flatten + conv

(4) 1x1x800 ---> conv

(5) 1x1x2(output)

Train:

RECG_CNN.py train

Predict

RECG_CNN.py predict

---

2. Extracting the truth part

Structure:

(1) (input)64x64x3 ---> conv + pool

(2) 32x32x32 ---> conv + pool

(3) 16x16x64 ---> transpose, stride = 2

(4-2) 32x32x32 ---> transpose, stride = 2

(5-1) 64x64x3(output)

Train:

CUTOUT_FCN.py

Predict:

pred = sess.run(annotation_pred,feed_dict)

pred = np.squeeze(pred,axis=3)

3. Others

Structure of 8layers:  

Train:

RECG_CNN.py train

CUTOUT_FCN.py train, need the 8layer model from RECG_CNN2.py.  

或

CUTOUT_segnet.py train

Predict:  

RECG_CNN.py predict

CUTOUT_FCN.py predict  

或

CUTOUT_segnet.py predict