diff --git a/README.md b/README.md index ebc6c39..d0e3e08 100644 --- a/README.md +++ b/README.md @@ -58,6 +58,8 @@ The saved features will take up approximately **29GB disk space** (for all image ### Training +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` + 1. Train with ground-truth layout (cloning expert + policy search after cloning) * Step a (cloning expert): `python exp_clevr/train_clevr_gt_layout.py` @@ -78,6 +80,8 @@ The downloaded snapshots should be placed under `exp_clevr/tfmodel/clevr_gt_layo ### Test +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` + 1. Evaluate *clevr_gt_layout* (cloning expert): `python exp_clevr/eval_clevr.py --exp_name clevr_gt_layout --snapshot_name 00050000 --test_split val` Expected accuracy: 78.9% (on val split). @@ -142,6 +146,7 @@ Note: this repository already contains the parsing results from Stanford Parser Train with ground-truth layout: +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` 1. Step a (cloning expert): `python exp_vqa/train_vqa_gt_layout.py` 2. Step b (policy search after cloning): @@ -156,6 +161,8 @@ The downloaded snapshots should be placed under `exp_vqa/tfmodel/vqa_gt_layout` ### Test +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` + 1. Evaluate on *vqa_gt_layout* (cloning expert): - (on test-dev2015 split): `python exp_vqa/eval_vqa.py --exp_name vqa_gt_layout --snapshot_name 00040000 --test_split test-dev2015` @@ -176,6 +183,8 @@ A copy of the SHAPES dataset is contained in this repository under `exp_shapes/s ### Training +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` + 1. Train with ground-truth layout (behavioral cloning from expert): `python exp_shapes/train_shapes_gt_layout.py` @@ -186,6 +195,8 @@ Note: by default, the above scripts use GPU 0. To train on a different GPU, set ### Test +0. Add the root of this repository to PYTHONPATH: `export PYTHONPATH=.:$PYTHONPATH` + 1. Evaluate *shapes_gt_layout* (behavioral cloning from expert): `python exp_shapes/eval_shapes.py --exp_name shapes_gt_layout --snapshot_name 00040000 --test_split test`