New eval

.gitignore

@@ -0,0 +1,5 @@
+i2s
+pascal3d_checkpoint.pt
+modelnet10so3_checkpoint.pt
+modelnet10_test.npz
+__pycache__/

Run in colab

@@ -0,0 +1,138 @@
+Run in colab
+
+
+# 1. Clone the repo
+!git clone https://github.com/Airport237/image2sphere.git
+
+# 2. Move into the folder
+%cd image2sphere
+
+# 3. Checkout the specific branch
+!git fetch origin
+!git checkout add-translation-head
+
+
+
+
+# !pwd (make sure in i2s)
+
+from google.colab import drive
+drive.mount('/content/drive')
+! unzip /content/drive/MyDrive/smallspeed.zip -d /content/speedplus_data
+
+# '''
+# Preprocess Images into correct format
+# Code adapted from https://github.com/tpark94/spnv2/blob/main/tools/preprocess.py
+# '''
+# import json
+# import os
+# import cv2
+# import tqdm
+# import numpy as np
+# import torch #added for sam2
+
+# datadir = '/content/speedplus_data/lightbox'
+# # Read labels from JSON file
+# jsonfile = '/content/speedplus_data/lightbox/test.json'
+# print(f'Reading JSON file from {jsonfile}...')
+# with open(jsonfile, 'r') as f:
+#     labels = json.load(f) # list
+
+# domain = ''
+# split = 'test.json'
+# outdir = os.path.join(datadir, domain, 'labels')
+# if not os.path.exists(outdir): os.makedirs(outdir)
+# csvfile = os.path.join(outdir, split.replace('json', 'csv'))
+# print(f'Label CSV file will be saved to {csvfile}')
+
+# # Where to save resized image?
+# imagedir = os.path.join(datadir, domain,
+#         f'images_{768}x{512}_RGB')
+# if not os.path.exists(imagedir): os.makedirs(imagedir)
+# print(f'Resized images will be saved to {imagedir}')
+
+
+# # if args.load_masks:
+# #     maskdir = os.path.join(datadir, domain,
+# #         f'masks_{int(cfg.DATASET.INPUT_SIZE[0]/cfg.DATASET.OUTPUT_SIZE[0])}x{int(cfg.DATASET.INPUT_SIZE[1]/cfg.DATASET.OUTPUT_SIZE[0])}')
+# #     if not os.path.exists(maskdir): os.makedirs(maskdir)
+# #     print(f'Resized masks will be saved to {maskdir}')
+
+# # Open
+# csv = open(csvfile, 'w')
+
+# for idx in tqdm.tqdm(range(len(labels))):
+
+#     # ---------- Read image & resize & save
+#     filename = labels[idx]['filename']
+#     image    = cv2.imread(os.path.join(datadir, 'images', filename), cv2.IMREAD_COLOR)
+#     image    = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+#     # image    = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
+#     image    = cv2.resize(image, [768, 512])
+#     cv2.imwrite(os.path.join(imagedir, filename), image)
+
+#     # ---------- Read mask & resize & save
+#     # if args.load_masks:
+#     #     mask     = cv2.imread(os.path.join(datadir, domain, 'masks', filename), cv2.IMREAD_GRAYSCALE)
+#     #     mask_out = cv2.resize(mask, [int(s / cfg.DATASET.OUTPUT_SIZE[0]) for s in cfg.DATASET.INPUT_SIZE])
+#     #     cv2.imwrite(os.path.join(maskdir, filename), mask_out)
+
+#     # ---------- Read labels
+#     q_vbs2tango = np.array(labels[idx]['q_vbs2tango_true'], dtype=np.float32)
+#     r_Vo2To_vbs = np.array(labels[idx]['r_Vo2To_vbs_true'], dtype=np.float32)
+
+#     # ---------- Bounding box labels
+#     # If masks are available, get them from masks
+#     # If not, use keypoints instead
+#     # if args.load_masks:
+#     #     seg  = np.where(mask > 0)
+#     #     xmin = np.min(seg[1]) / camera['Nu']
+#     #     ymin = np.min(seg[0]) / camera['Nv']
+#     #     xmax = np.max(seg[1]) / camera['Nu']
+#     #     ymax = np.max(seg[0]) / camera['Nv']
+
+
+#     # CSV row
+#     row = [filename]
+
+#     row = row + q_vbs2tango.tolist() \
+#           + r_Vo2To_vbs.tolist() \
+#           # + [xmin, ymin, xmax, ymax]
+#     row = ', '.join([str(e) for e in row])
+
+#     # Write
+#     csv.write(row + '\n')
+
+# csv.close()
+!pip install -r requirements.txt #gonna fail
+!pip install e3nn
+!pip install healpy
+
+%cd /content/image2sphere
+
+%cd /content/image2sphere
+
+!python -m src.train \
+  --dataset_name speed+ \
+  --dataset_path /content/speedplus_data/ \
+  --results_dir /content/results_speedplus \
+  --device cuda \
+  --num_epochs 5 \
+  --batch_size 2
+'''
+%cd /content/image2sphere
+
+!python -m src.train \
+  --dataset_name speed+ \
+  --dataset_path /content/speedplus_data \
+  --results_dir /content/results_speedplus_lowmem \
+  --device cpu \
+  --num_epochs 1 \
+  --batch_size 1 \
+  --num_workers 0 \
+  --encoder resnet18_pretrained \
+  --sphere_fdim 512 \
+  --train_grid_n_points 4096 \
+  --train_grid_rec_level 3 \
+  --eval_grid_rec_level 5
+'''